Pith. sign in

REVIEW 3 major objections 3 minor 50 references

Model Accuracy and Data Heterogeneity Shape Uncertainty Quantification in Machine Learning Interatomic Potentials

T0 review · 3 major / 3 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 The paper claims that uncertainty quantification reliability in machine-learned interatomic potentials is governed by model accuracy and data homogeneity, and that clustering-enhanced local D-optimality improves novelty detection on…

desk verdict The supplied full text is a different paper (routing problems), so the UQ claims rest on an unverifiable abstract that alone does not justify peer review. read the letter →

arxiv 2508.03405 v1 pith:PKRS45AD submitted 2025-08-05 cond-mat.mtrl-sci cs.LG

Fei Shuang , Zixiong Wei , Kai Liu , Wei Gao , Poulumi Dey This is my paper
classification cond-mat.mtrl-scics.LG
keywords machinelearninginteratomicpotentialsuncertaintyquantificationensembleD-optimalityatomicclusterexpansionnoveltydetectionactivedataheterogeneity
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 studies how reliable uncertainty quantification (UQ) is for machine-learned interatomic potentials, focusing on two strategies: ensemble learning and D-optimality. It argues that both strategies give well-calibrated uncertainty estimates on homogeneous training sets, but on heterogeneous datasets they understate errors and lose sensitivity to novel atomic environments. The central claim is that higher model accuracy strengthens the correlation between predicted uncertainties and actual errors and improves novelty detection, while D-optimality produces more conservative estimates. To fix the heterogeneity problem, the authors propose clustering-enhanced local D-optimality, which partitions configuration space during training and applies D-optimality within each cluster, and they report that this approach substantially improves the detection of novel atomic environments in heterogeneous data. If correct, these findings clarify how model fidelity and data homogeneity govern UQ performance and point toward more robust active learning and adaptive sampling for interatomic potential development.

What carries the argument

The central objects are two UQ strategies operating inside the atomic cluster expansion (ACE) framework: ensemble learning, which takes the spread of predictions across several trained models as the uncertainty, and D-optimality, a design criterion that selects configurations so as to maximize the determinant of the Fisher information matrix, used here to flag configurations where the model is uncertain. The paper's new mechanism is clustering-enhanced local D-optimality: the configuration space is partitioned into clusters during training, and D-optimality is applied separately within each cluster. This makes novelty judgments local, comparing a new atomic environment with the training density in its own region of configuration space rather than against a global average.

What would settle it

A controlled comparison that fixes a heterogeneous dataset and varies the clustering algorithm and the number of clusters, measuring novelty-detection performance against a global D-optimality baseline, would settle whether the gain is intrinsic to local D-optimality or an artifact of a particular partition choice.

Watch

Extended reading notes

Core claim

Within the atomic cluster expansion (ACE) framework, the paper establishes that the quality of uncertainty quantification is controlled by two factors: model fidelity and data homogeneity. Higher model accuracy increases the correlation between a predicted uncertainty and the actual error, and it also improves the model's ability to flag novel atomic environments; D-optimality tends to give more conservative uncertainty estimates than ensemble learning. On homogeneous training sets, both methods produce well-calibrated uncertainties, but on heterogeneous sets they underpredict errors and show reduced novelty sensitivity. The introduced clustering-enhanced local D-optimality method partitions configuration space into clusters during training and applies D-optimality within each cluster, and the authors find that this local scheme substantially improves detection of novel atomic environments in heterogeneous datasets. The paper presents this method as a practical route to robust active learning and adaptive sampling in MLIP development.

Load-bearing premise

The method's benefit on heterogeneous data rests on the assumption that configuration space can be partitioned into meaningful clusters during training; the paper does not specify the clustering algorithm or the number of clusters, so the reported improvement could depend on those unspecified choices.

Editorial extensions

If this is right

  • On homogeneous training data, both ensemble learning and D-optimality give well-calibrated uncertainty estimates, so practitioners can trust these numbers for interpolation within well-sampled regions.
  • Because D-optimality produces more conservative uncertainty estimates than ensembles, safety-conscious applications may prefer it despite the wider error bars.
  • On heterogeneous training sets, standard UQ underpredicts errors and loses novelty sensitivity, which means active learning loops that use these estimates will tend to miss the very environments that need new data.
  • The clustering-enhanced local D-optimality scheme restores novelty detection on heterogeneous data, giving adaptive sampling a concrete tool for uneven training sets.
  • Higher model accuracy directly strengthens UQ reliability, so improving model fidelity is also a way to make uncertainty estimates more trustworthy.

Reading between the lines

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

  • The paper's results suggest that UQ reliability is limited less by the choice between ensemble and D-optimality than by the match between training data and target distribution; any method that assumes a single global error model may be inherently fragile on heterogeneous datasets.
  • A natural test of the mechanism would be to apply cluster-local D-optimality to other machine-learned interatomic potential architectures, such as neural network potentials; the improvement should transfer if the benefit comes from localizing the design criterion rather than from ACE-specific features.
  • The cluster partition itself could be reused as an interpretable segmentation of the potential energy surface, giving users a map of where the model is expected to be reliable.
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

3 major / 3 minor

Summary. The manuscript under review, arXiv:2508.03405, claims to investigate two uncertainty quantification (UQ) strategies for machine learning interatomic potentials (MLIPs) within the atomic cluster expansion (ACE) framework: ensemble learning and D-optimality. The abstract states that higher model accuracy strengthens the correlation between predicted uncertainties and actual errors, that D-optimality gives more conservative estimates, that both methods are well calibrated on homogeneous data but underpredict errors on heterogeneous data, and that a proposed clustering-enhanced local D-optimality substantially improves novelty detection on heterogeneous datasets. However, the full text supplied for this manuscript is actually arXiv:2508.03406, a paper titled "Multi-Objective Infeasibility Diagnosis for Routing Problems Using Large Language Models," which is explicitly labeled in its appendix as an appendix for a different paper. This supplied full text contains no content on MLIPs, ACE, ensemble learning, D-optimality, calibration, or clustering-enhanced local D-optimality. Consequently, the central claims of the target manuscript are not supported by any derivation or experimental evidence in the provided material.

Significance. If the claims in the abstract were substantiated, the work could make a meaningful contribution to uncertainty quantification for MLIPs, particularly for active learning and adaptive sampling on heterogeneous training sets. The abstract articulates a clear, falsifiable hypothesis about the roles of model fidelity and data heterogeneity, and the proposed clustering-enhanced local D-optimality is a potentially useful methodological idea. However, because the supplied full text corresponds to a different paper entirely, the technical content behind these claims is completely unavailable to this review. The significance of the work therefore cannot be assessed beyond the abstract's assertions; the manuscript in its current review form does not permit evaluation of the derivation, experiments, or reproducibility artifacts.

major comments (3)
  1. [Full text (supplied as arXiv:2508.03406)] The full text provided for this submission is not the manuscript of arXiv:2508.03405. It is titled "Multi-Objective Infeasibility Diagnosis for Routing Problems Using Large Language Models" and its appendix explicitly states, "This is an appendix for 'Multi-Objective Infeasibility Diagnosis for Routing Problems Using Large Language Models'." None of its sections, equations, or tables refer to atomic cluster expansion, ensemble learning, D-optimality, calibration, novelty detection, or clustering-enhanced local D-optimality. This is not a minor formatting issue: without the correct full text, every quantitative claim in the abstract is untestable, and the manuscript's central derivation and experimental support are entirely missing from this review package.
  2. [Abstract (empirical claims)] The abstract makes strong empirical statements such as "Both methods deliver well calibrated uncertainties on homogeneous training sets, yet they underpredict errors and exhibit reduced novelty sensitivity on heterogeneous datasets" and "This approach substantially improves the detection of novel atomic environments." No quantitative results, dataset descriptions, error bars, statistical tests, or effect sizes are provided anywhere in the available material. Even if the correct full text were supplied, the abstract alone does not allow a reader to assess the magnitude of the claimed effects or the calibration metric used. These load-bearing empirical claims need specific numerical evidence in the main text.
  3. [Abstract (method specification)] The proposed clustering-enhanced local D-optimality is described only in one sentence: "it partitions configuration space into clusters during training and applies D-optimality within each cluster." The clustering algorithm, the number of clusters, the feature representation on which clustering is performed, and the precise definition of local D-optimality are not specified. The effectiveness of the method could depend critically on these choices, and their absence makes the method irreproducible on the basis of the abstract alone. A full manuscript should provide these details, as well as a sensitivity analysis with respect to the number of clusters.
minor comments (3)
  1. [Abstract (terminology)] The term "well calibrated" is used without defining the calibration metric (e.g., reliability diagram, expected calibration error, or coverage probability) or the acceptance threshold; this should be specified in the main text.
  2. [Abstract (datasets)] The abstract does not mention which datasets, materials, or ACE potentials were used; such experimental context is necessary for interpreting the claimed differences between homogeneous and heterogeneous training sets.
  3. [Abstract (clarity of claims)] The phrase "higher model accuracy strengthens the correlation between predicted uncertainties and actual errors" is ambiguous about whether this is an observed correlation or a causal claim controlled through model architecture or training set size; the main text should clarify the experimental design.

Circularity Check

0 steps flagged · score 0.0 of 10

Supplied full text is the wrong paper; no quotable circular step can be established from the target abstract alone.

full rationale

The target manuscript is arXiv:2508.03405, whose abstract claims that 'higher model accuracy strengthens the correlation between predicted uncertainties and actual errors and improves novelty detection' and that 'clustering-enhanced local D-optimality... substantially improves the detection of novel atomic environments.' The supplied full text, however, is arXiv:2508.03406v1, 'Multi-Objective Infeasibility Diagnosis for Routing Problems Using Large Language Models,' and it explicitly states at its appendix header: 'This is an appendix for "Multi-Objective Infeasibility Diagnosis for Routing Problems Using Large Language Models".' None of the target paper's equations, ACE framework, ensemble learning steps, D-optimality definitions, calibration metrics, or clustering procedure appears in the supplied evidence. Therefore there is no derivation chain to walk and no way to exhibit a specific reduction of a predicted quantity to a fitted input or to a self-citation. Under the hard rule that circularity must be demonstrated by exact quotes and equation identities, the honest finding is no circular step identified from the supplied text. This is a verification gap caused by missing full text, not an affirmative circularity verdict; the score reflects the absence of quotable circularity rather than validation of the paper's content.

Assumptions & free parameters 1 free parameters · 2 assumptions · 0 invented entities

The abstract is the only available content; the full text belongs to a different paper. Therefore the ledger below is necessarily tentative, inferred from the abstract's claims. No specific fitted values or entities can be identified without the actual manuscript.

free parameters (1)
  • Clustering hyperparameters (e.g., number of clusters)
    The abstract does not disclose how clusters are formed or how many are used; these are likely user-specified or fitted and affect the UQ novelty detection performance.
assumptions (2)
  • domain assumption D-optimality in the atomic cluster expansion feature space is a valid proxy for predictive uncertainty in MLIPs.
    The abstract claims D-optimality produces conservative uncertainty estimates that correlate with errors; this relies on the optimal-design interpretation of the ACE information matrix, which is not proven in the abstract.
  • ad hoc to paper Clustering configuration space during training yields local regions within which D-optimality is meaningful.
    The clustering-enhanced local D-optimality method is introduced but its clustering criterion and validity are unspecified; this is a load-bearing modeling choice that could be an artifact if clusters do not align with error structure.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Model Accuracy and Data Heterogeneity Shape Uncertainty Quantification in Machine Learning Interatomic Potentials." pith.science (2026). https://pith.science/paper/PKRS45AD

@misc{pith2026250803405,
  author       = {Pith},
  title        = {Pith review of: Model Accuracy and Data Heterogeneity Shape Uncertainty Quantification in Machine Learning Interatomic Potentials},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PKRS45AD}},
  note         = {Machine review of arXiv:2508.03405}
}
read the original abstract

Machine learning interatomic potentials (MLIPs) enable accurate atomistic modelling, but reliable uncertainty quantification (UQ) remains elusive. In this study, we investigate two UQ strategies, ensemble learning and D-optimality, within the atomic cluster expansion framework. It is revealed that higher model accuracy strengthens the correlation between predicted uncertainties and actual errors and improves novelty detection, with D-optimality yielding more conservative estimates. Both methods deliver well calibrated uncertainties on homogeneous training sets, yet they underpredict errors and exhibit reduced novelty sensitivity on heterogeneous datasets. To address this limitation, we introduce clustering-enhanced local D-optimality, which partitions configuration space into clusters during training and applies D-optimality within each cluster. This approach substantially improves the detection of novel atomic environments in heterogeneous datasets. Our findings clarify the roles of model fidelity and data heterogeneity in UQ performance and provide a practical route to robust active learning and adaptive sampling strategies for MLIP development.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 23 canonical work pages

  1. [1]

    Friederich, F

    P. Friederich, F. H \"a se, J. Proppe, A. Aspuru-Guzik, Machine-learned potentials for next-generation matter simulations. Nature Materials 20 (6), 750--761 (2021)

    work page 2021

  2. [2]

    Ceriotti, Beyond potentials: Integrated machine learning models for materials

    M. Ceriotti, Beyond potentials: Integrated machine learning models for materials. MRS Bulletin 47 (10), 1045--1053 (2022)

    work page 2022

  3. [3]

    Zuo, et al., Performance and Cost Assessment of Machine Learning Interatomic Potentials

    Y. Zuo, et al., Performance and Cost Assessment of Machine Learning Interatomic Potentials. The Journal of Physical Chemistry A 124, 731--745 (2020), doi:10.1021/acs.jpca.9b08723, https://pubs.acs.org/doi/10.1021/acs.jpca.9b08723

    doi:10.1021/acs.jpca.9b08723 2020

  4. [4]

    Jacobs, et al., A practical guide to machine learning interatomic potentials--Status and future

    R. Jacobs, et al., A practical guide to machine learning interatomic potentials--Status and future. Current Opinion in Solid State and Materials Science 35, 101214 (2025)

    work page 2025

  5. [5]

    Qamar, M

    M. Qamar, M. Mrovec, Y. Lysogorskiy, A. Bochkarev, R. Drautz, Atomic Cluster Expansion for Quantum-Accurate Large-Scale Simulations of Carbon. Journal of Chemical Theory and Computation 19, 5151--5167 (2023), doi:10.1021/acs.jctc.2c01149, https://pubs.acs.org/doi/10.1021/acs.jctc.2c01149

    doi:10.1021/acs.jctc.2c01149 2023

  6. [6]

    Liang, et al., Atomic cluster expansion for Pt--Rh catalysts: From ab initio to the simulation of nanoclusters in few steps

    Y. Liang, et al., Atomic cluster expansion for Pt--Rh catalysts: From ab initio to the simulation of nanoclusters in few steps. Journal of Materials Research 38 (24), 5125--5135 (2023)

    work page 2023

  7. [7]

    L. C. Erhard, J. Rohrer, K. Albe, V. L. Deringer, Modelling atomic and nanoscale structure in the silicon–oxygen system through active machine learning. Nature Communications 15, 1927 (2024), doi:10.1038/s41467-024-45840-9, https://www.nature.com/articles/s41467-024-45840-9

    doi:10.1038/s41467-024-45840-9 2024

  8. [8]

    A. V. Shapeev, Moment Tensor Potentials: A Class of Systematically Improvable Interatomic Potentials. Multiscale Modeling & Simulation 14, 1153--1173 (2016), doi:10.1137/15M1054183, http://epubs.siam.org/doi/10.1137/15M1054183

    doi:10.1137/15m1054183 2016

Show all 50 references
  1. [9]

    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 Technology 2, 025002 (2021), doi:10.1088/2632-2153/abc9fe, https://iopscience.iop.org/article/10.1088/2632-2153/abc9fe

    2021 doi

  2. [10]

    Podryabinkin, K

    E. Podryabinkin, K. Garifullin, A. Shapeev, I. Novikov, MLIP-3: Active learning on atomic environments with moment tensor potentials. The Journal of Chemical Physics 159, 84112 (2023), doi:10.1063/5.0155887, https://pubs.aip.org/jcp/article/159/8/084112/2908187/MLIP-3-Active-l...

    2023 doi

  3. [11]

    Drautz, Atomic cluster expansion for accurate and transferable interatomic potentials

    R. Drautz, Atomic cluster expansion for accurate and transferable interatomic potentials. Physical Review B 99, 014104 (2019), doi:10.1103/PhysRevB.99.014104, https://link.aps.org/doi/10.1103/PhysRevB.99.014104

    2019 doi

  4. [12]

    Lysogorskiy, et al., Performant implementation of the atomic cluster expansion (PACE) and application to copper and silicon

    Y. Lysogorskiy, et al., Performant implementation of the atomic cluster expansion (PACE) and application to copper and silicon. npj Computational Materials 7, 97 (2021), doi:10.1038/s41524-021-00559-9, https://www.nature.com/articles/s41524-021-00559-9

    2021 doi

  5. [13]

    Bochkarev, et al., Efficient parametrization of the atomic cluster expansion

    A. Bochkarev, et al., Efficient parametrization of the atomic cluster expansion. Physical Review Materials 6, 013804 (2022), doi:10.1103/PhysRevMaterials.6.013804, https://link.aps.org/doi/10.1103/PhysRevMaterials.6.013804

    2022 doi

  6. [14]

    Xu, et al., GPUMD 4.0: A high‐performance molecular dynamics package for versatile materials simulations with machine‐learned potentials

    K. Xu, et al., GPUMD 4.0: A high‐performance molecular dynamics package for versatile materials simulations with machine‐learned potentials. Materials Genome Engineering Advances (2025), doi:10.1002/mgea.70028, http://dx.doi.org/10.1002/mgea.70028

    2025 doi

  7. [15]

    \'A . D. Carral, et al., Stability of binary precipitates in Cu-Ni-Si-Cr alloys investigated through active learning. Materials Chemistry and Physics 306, 128053 (2023)

    2023

  8. [16]

    X. Xu, X. Zhang, E. Bitzek, S. Schmauder, B. Grabowski, Origin of the yield stress anomaly in L12 intermetallics unveiled with physically informed machine-learning potentials. Acta Materialia 281, 120423 (2024)

    2024

  9. [17]

    Rybin, I

    N. Rybin, I. S. Novikov, A. Shapeev, Accelerating structure prediction of molecular crystals using actively trained moment tensor potential. Physical Chemistry Chemical Physics 27 (10), 5141--5148 (2025)

    2025

  10. [18]

    Klimanova, N

    O. Klimanova, N. Rybin, A. Shapeev, Accelerating the global search of adsorbate molecule positions using machine-learning interatomic potentials with active learning. Physical Chemistry Chemical Physics 27 (17), 9201--9210 (2025)

    2025

  11. [19]

    A. S. Kotykhov, et al., Actively trained magnetic moment tensor potentials for mechanical, dynamical, and thermal properties of paramagnetic CrN. Physical Review B 111 (9), 094438 (2025)

    2025

  12. [20]

    van der Oord, M

    C. van der Oord, M. Sachs, D. P. Kov \'a cs, C. Ortner, G. Cs \'a nyi, Hyperactive learning for data-driven interatomic potentials. npj Computational Materials 9 (1), 168 (2023)

    2023

  13. [21]

    Perez, A

    D. Perez, A. Subramanyam, I. Maliyov, T. D. Swinburne, Uncertainty quantification for misspecified machine learned interatomic potentials. arXiv preprint arXiv:2502.07104 (2025)

    2025 arXiv

  14. [22]

    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

  15. [23]

    Lysogorskiy, A

    Y. Lysogorskiy, A. Bochkarev, M. Mrovec, R. Drautz, Active learning strategies for atomic cluster expansion models. Physical Review Materials 7, 043801 (2023), doi:10.1103/PhysRevMaterials.7.043801, https://link.aps.org/doi/10.1103/PhysRevMaterials.7.043801

    2023 doi

  16. [24]

    Shuang, et al., Modeling extensive defects in metals through classical potential-guided sampling and automated configuration reconstruction

    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

  17. [25]

    Batatia, D

    I. Batatia, D. P. Kovacs, G. Simm, C. Ortner, G. Cs \'a nyi, MACE: Higher order equivariant message passing neural networks for fast and accurate force fields. Advances in Neural Information Processing Systems 35, 11423--11436 (2022)

    2022

  18. [26]

    Zhang, G

    L. Zhang, G. Csányi, E. van der Giessen, F. Maresca, Atomistic fracture in bcc iron revealed by active learning of Gaussian approximation potential. npj Computational Materials 9 (1) (2023), doi:10.1038/s41524-023-01174-6, http://dx.doi.org/10.1038/s41524-023-01174-6

    2023 doi

  19. [27]

    Byggm \"a star, K

    J. Byggm \"a star, K. Nordlund, F. Djurabekova, Gaussian approximation potentials for body-centered-cubic transition metals. Physical Review Materials 4 (9), 093802 (2020)

    2020

  20. [28]

    Liang, et al., NEP89: Universal neuroevolution potential for inorganic and organic materials across 89 elements

    T. Liang, et al., NEP89: Universal neuroevolution potential for inorganic and organic materials across 89 elements. arXiv preprint arXiv:2504.21286 (2025)

    2025 arXiv

  21. [29]

    M. Poul, L. Huber, J. Neugebauer, Automated generation of structure datasets for machine learning potentials and alloys. npj Computational Materials 11 (1), 174 (2025)

    2025

  22. [30]

    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, 104103 (2023), doi:10.1103/PhysRevB.107.104103, https://link.aps.org/doi/10.1103/PhysRevB.107.104103

    2023 doi

  23. [31]

    Hodapp, A

    M. Hodapp, A. Shapeev, In operando active learning of interatomic interaction during large-scale simulations. Machine Learning: Science and Technology 1, 045005 (2020), doi:10.1088/2632-2153/aba373, https://iopscience.iop.org/article/10.1088/2632-2153/aba373

    2020 doi

  24. [32]

    Mismetti, M

    L. Mismetti, M. Hodapp, Automated atomistic simulations of dissociated dislocations with ab initio accuracy. Physical Review B 109, 094120 (2024), doi:10.1103/PhysRevB.109.094120, https://link.aps.org/doi/10.1103/PhysRevB.109.094120

    2024 doi

  25. [33]

    J. Qi, T. W. Ko, B. C. Wood, T. A. Pham, S. P. Ong, Robust training of machine learning interatomic potentials with dimensionality reduction and stratified sampling. npj Computational Materials 10, 43 (2024), doi:10.1038/s41524-024-01227-4, https://www.nature.com/articles/s415...

    2024 doi

  26. [34]

    Schwalbe-Koda, S

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

    2025

  27. [35]

    A. P. Bartók, R. Kondor, G. Csányi, On representing chemical environments. Physical Review B 87, 184115 (2013), doi:10.1103/PhysRevB.87.184115, https://link.aps.org/doi/10.1103/PhysRevB.87.184115

    2013 doi

  28. [36]

    Batatia, et al., A foundation model for atomistic materials chemistry

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

    2023 arXiv

  29. [37]

    Batatia, et al., The design space of E (3)-equivariant atom-centred interatomic potentials

    I. Batatia, et al., The design space of E (3)-equivariant atom-centred interatomic potentials. Nature Machine Intelligence 7 (1), 56--67 (2025)

    2025

  30. [38]

    Drautz, Atomic cluster expansion of scalar, vectorial, and tensorial properties including magnetism and charge transfer

    R. Drautz, Atomic cluster expansion of scalar, vectorial, and tensorial properties including magnetism and charge transfer. Physical Review B 102 (2), 024104 (2020)

    2020

  31. [39]

    Dusson, et al., Atomic cluster expansion: Completeness, efficiency and stability

    G. Dusson, et al., Atomic cluster expansion: Completeness, efficiency and stability. Journal of Computational Physics 454, 110946 (2022), doi:10.1016/j.jcp.2022.110946, https://linkinghub.elsevier.com/retrieve/pii/S0021999122000080

    2022

  32. [40]

    Thompson, L

    A. Thompson, L. Swiler, C. Trott, S. Foiles, G. Tucker, Spectral neighbor analysis method for automated generation of quantum-accurate interatomic potentials. Journal of Computational Physics 285, 316--330 (2015), doi:10.1016/j.jcp.2014.12.018, https://linkinghub.elsevier.com/...

    2015 doi

  33. [41]

    A. H. Larsen, et al., The atomic simulation environment—a Python library for working with atoms. Journal of Physics: Condensed Matter 29 (27), 273002 (2017)

    2017

  34. [42]

    A. P. Thompson, et al., LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales. Computer Physics Communications 271, 108171 (2022), doi:10.1016/j.cpc.2021.108171, https://linkinghub.elsevier.com/retrieve/pii/S0010465...

    2022

  35. [43]

    Kresse, J

    G. Kresse, J. Furthmüller, Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Physical Review B 54, 11169--11186 (1996), doi:10.1103/PhysRevB.54.11169, https://link.aps.org/doi/10.1103/PhysRevB.54.11169

    1996 doi

  36. [44]

    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, https://link.aps.org/doi/10.1103/PhysRevLett.77.3865

    1996 doi

  37. [45]

    P. E. Bl \"o chl, Projector augmented-wave method. Physical Review B 50 (24), 17953 (1994)

    1994

  38. [46]

    V. Wang, N. Xu, J.-C. Liu, G. Tang, W.-T. Geng, VASPKIT: A user-friendly interface facilitating high-throughput computing and analysis using VASP code. Computer Physics Communications 267, 108033 (2021)

    2021

  39. [47]

    H. J. Monkhorst, J. D. Pack, Special points for Brillouin-zone integrations. Physical Review B 13 (12), 5188 (1976)

    1976

  40. [48]

    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, 015012 (2010), doi:10.1088/0965-0393/18/1/015012, https://iopscience.iop.org/article/10.1088/0965-039...

    2010 doi

  41. [49]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key month note number organization pages publisher school series title type url volume year label INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION in...

  42. [50]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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