Pith. sign in

REVIEW 3 major objections 4 minor 66 references

Dynamical Data for More Efficient and Generalizable Learning: A Case Study in Disordered Elastic Networks

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Training a graph-neural-network simulator on raw compression dynamics alone lets it reproduce the motion of unseen disordered elastic networks and predict their emergent Poisson's ratio, generalizing well beyond the training distribution.

desk verdict Solid dynamics learning case study whose headline property claims rest on an unvalidated Poisson estimator; send to review with conditions. read the letter →

arxiv 2505.21125 v2 pith:GJFOML5M submitted 2025-05-27 physics.chem-ph cond-mat.mtrl-sci

Salman N. Salman , Sergey A. Shteingolts , Ron Levie , Dan Mendels This is my paper
classification physics.chem-phcond-mat.mtrl-sci
keywords graphneuralnetworksimulatordisorderedelasticnetworksdynamicaldataPoisson'sratioout-of-distributiongeneralizationefficiencymoleculardynamicsemergentproperties
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 tests whether dynamical data can replace large static, property-labeled datasets in materials machine learning. The authors train a graph-neural-network simulator on raw compression trajectories of two-dimensional disordered elastic networks and find that, from a small number of example systems, it accurately reproduces the motion of unseen networks. The simulator predicts the emergent Poisson's ratio and its strain dependence without explicit supervision, and it generalizes across temperature, strain amplitude, and Poisson's ratios beyond the training range. The authors argue that this makes dynamical data a route to data-efficient and generalizable materials design, especially where data acquisition is costly.

What carries the argument

The central machinery is a graph neural network with an encoder–processor–decoder architecture that operates on graphs whose nodes are particles and whose edges are harmonic bonds. Node features are velocity histories over up to three past steps, and edge features are the bond vector, its length, and a stiffness factor. The processor applies several message-passing layers, and the decoder outputs per-node accelerations, which are advanced by forward Euler integration so that full compression rollouts are generated autoregressively. Training minimizes only the mean-squared error between predicted and ground-truth accelerations, so property predictions such as Poisson's ratio are emergent rather than explicitly supervised.

What would settle it

Recompute the Poisson's ratios of the out-of-distribution test networks from their elastic moduli, $C_{11}$ and $C_{12}$ averaged over directions as described in the supplementary material, and compare those values against the displacement-based estimates used in the paper's parity plots; if the two disagree systematically, especially for auxetic networks, the claimed generalization would be an artifact of the estimator rather than a true property prediction.

Watch

Extended reading notes

Core claim

The central claim is that a graph-neural-network simulator can learn the physical dynamics of uniaxial compression in two-dimensional disordered elastic networks from a small number of example trajectories and then accurately reproduce the temporal evolution of unseen networks. The authors also show that a system-level elastic property, the Poisson's ratio, and its strain dependence emerge correctly from the learned dynamics even though the model is not trained to predict them. The simulator further extrapolates beyond its training distribution: it handles temperature changes, strain amplitudes up to 20 times larger than trained on, networks with roughly twice the node count, and Poisson's ratios outside the training range. Generalization is strongest when training on highly auxetic (negative Poisson's ratio) networks, whose more complex and heterogeneous dynamics transfer to simpler non-auxetic regimes more readily than in the opposite direction.

Load-bearing premise

The claim that the model generalizes to out-of-distribution Poisson's ratios rests on a custom displacement-based estimator, $\nu_{\rm est} = -D_y/D_x$, that the paper itself describes as 'not strictly accurate' with errors up to 5%, and the same estimator is used both as the ground truth in parity tests and to define the training-range partition, so a systematic bias for auxetic or optimized networks could make the reported generalization an artifact.

Editorial extensions

If this is right

  • A simulator trained on only a small fraction of the 3,500 generated trajectories matches the accuracy of the full model, so the marginal cost of new system-to-property predictions drops sharply.
  • Properties that are difficult to label directly, such as Poisson's ratio and its strain dependence, can be read off from learned dynamics instead of being trained with explicit supervision.
  • Training on highly auxetic (negative-Poisson-ratio) dynamics transfers to non-auxetic systems, whereas the reverse direction is more limited; the more complex regime covers the simpler one.
  • The simulator extends to networks about twice the trained node count and to strain amplitudes up to 20 times larger than those seen in training.

Reading between the lines

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

  • If the data-efficiency result transfers, the practical recipe is: collect a few long dynamical trajectories from a few representative systems, train an autoregressive graph simulator, and read off any equilibrium or transport property observable in the dynamics, without building a large property-labeled database.
  • The asymmetric generalization suggests a curriculum: include the most heterogeneous or complex dynamical regimes in the training set, and simpler regimes are likely to be covered as a byproduct.
  • Because the parity plots use the same displacement-based estimator for predicted and ground-truth Poisson's ratios, the reported agreement could partly reflect shared estimator bias; recomputing ground truth from elastic moduli, as the supplementary material describes, would provide a stricter test of whether the model genuinely learns the elastic property.
  • The strain-amplitude generalization is plausibly a sign that the model has learned the linear-response regime; deformations that trigger bond breaking or plastic rearrangements would likely break the extrapolation, so the 20x strain claim should be seen as valid within the harmonic regime unless tested further.
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 / 4 minor

Summary. The paper proposes a graph neural network (GNN) simulator trained on uniaxial compression trajectories of two-dimensional disordered elastic networks. The model predicts accelerations from velocity and bond features and is rolled out autoregressively. The authors report low position MSE on unseen network dynamics, accurate prediction of Poisson's ratio as an emergent property from a small number of training trajectories, and out-of-distribution generalization across temperature, strain amplitude, and Poisson's ratio. The central claim is that dynamical data can replace large property-labeled datasets for system-to-property learning and support extrapolation beyond the training distribution.

Significance. The direct dynamics reproduction evidence is strong: single-step position MSE around 1e-10 and 1e-5 after 100 rollout steps, with clear overlay figures, and the data-efficiency trend in Fig. 5 is plausible. If the emergent-property and out-of-distribution claims were verified against a reliable Poisson's ratio measure, this would be an important contribution to data-efficient materials ML. The current manuscript, however, evaluates these claims with an unvalidated displacement-based estimator, which limits the significance until that point is addressed.

major comments (3)
  1. [Supplementary Sec. I.A.6 / Figs. 3, 5, 6, 7] Supplementary Sec. I.A.6 defines the Poisson's ratio estimator ν_est = -D_y/D_x and states that it is 'not strictly accurate' with errors up to 5%. This estimator is used as the ground truth in the parity plots (Fig. 3c,d), the data-efficiency curve (Fig. 5), the temperature generalization plot (Fig. 6), and the out-of-distribution Poisson's ratio plot (Fig. 7), yet it is never compared with the elastic-moduli-based Poisson's ratio from Supplementary Sec. I.A.5 that defines the training ranges. The parity plots therefore only demonstrate agreement between the predicted and ground-truth versions of the same proxy, not prediction of the true material property. The claimed OOD generalization to Poisson's ratios beyond the training range is not established unless the estimator is validated on the test set, especially for auxetic networks where the dynamics are most complex (Fig. 4 and Sec. III.B).
  2. [Supplementary Sec. I.A.6] The estimator in Supplementary Sec. I.A.6 uses absolute distances from the box center (|r_η|) to approximate the box strain. This quantity is sensitive to non-affine node displacements, which are prominent in auxetic networks (Fig. 4). The asserted 5% error bound is not derived or tested against the elastic-moduli definition, so the bias could be significantly larger for the most challenging test cases, precisely those that determine the OOD conclusions. A concrete test would be to compute both definitions of ν on a sample of ground-truth trajectories and report the discrepancy as a function of the true ν.
  3. [Fig. 7] The OOD analysis in Fig. 7 uses training-range assignments based on elastic-moduli ν (Sec. I.A.5) but evaluates performance with the displacement-based ν_est (Sec. I.A.6). If ν_est is a biased estimator of ν, the systems described as 'outside the training range' may not be outside the range in the metric used for scoring. The authors should clarify which ν definition is used to split the data, and ideally use the same definition consistently for training-range labeling and evaluation.
minor comments (4)
  1. [Sec. II] In the message-passing equation in Sec. II, there is a missing closing parenthesis after the sum over neighbors; the formula should read x_i^(k) = ψ^(k)( x_i^(k-1), ∑_{j∈N(i)} φ^(k)( x_i^(k-1), x_j^(k-1), e_ij) ).
  2. [Sec. III.B] The complexity measure α in Sec. III.B is defined with a sum from i=0 to N; it should likely run from i=1 to N.
  3. [Sec. III.C] The claims of generalization to networks twice the size and to strains 20 times larger than training (Sec. III.C) are stated without a supporting figure or quantitative metric; adding such evidence would strengthen the paper.
  4. [Fig. 3 and Fig. 7] The figure captions for Fig. 3 and Fig. 7 would be easier to interpret if they stated the number of test trajectories and the exact definition of the reported error (e.g., mean absolute error vs. root-mean-square).

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: Poisson's-ratio estimates are post-hoc functions of predicted dynamics, and no load-bearing claim reduces to a fitted parameter or self-citation.

full rationale

The derivation chain is self-contained. The simulator is trained to minimize MSE between predicted and ground-truth accelerations from LAMMPS trajectories (Sec. II, Implementation and training details); Poisson's ratio never appears in the loss or as a label. The emergent-property evaluation computes ν_est = -Dy/Dx (SI I.A.6) from rollout positions and compares predicted vs. ground-truth trajectories, so agreement in ν_est is a consequence of accurate dynamics rather than an input fitted to the property. The architecture is taken from external GNS references (refs. 40,41), and no load-bearing argument reduces to a self-citation by the present authors. The only flagged limitation is SI I.A.6's admission that ν_est is 'not strictly accurate' with errors up to 5%, and the fact that the training-range partitions use the elastic-moduli definition ν = (1-G/B)/(1+G/B) from SI I.A.5 while the evaluation uses ν_est; this is a measurement-validity mismatch, not a circular reduction, because the estimator is not calibrated to the elastic-moduli labels and the model is not trained on either definition. Accordingly, no step in the paper's derivation is equivalent to its inputs by construction.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the LAMMPS force field as ground truth, on the custom displacement-based Poisson estimator used in all evaluations, on the sufficiency of the chosen graph representation and integrator, and on the representativeness of the optimized OOD networks. The only hand-tuned quantities are standard architecture hyperparameters; no new physical entity is introduced.

free parameters (4)
  • history_steps_h = 3 (values 1 to 3 tested; caption mentions 2,3,4 inputs)
    Number of past velocity configurations used as node features; authors report improved rollout accuracy up to h=3 with diminishing returns.
  • message_passing_layers_k = not explicitly stated; 'more than one layer yields diminishing returns'
    Processor depth tested manually; no single value is fixed in the text.
  • hidden_dimension = 128
    Width of all MLP hidden layers; a design choice not fitted to physical data.
  • learning_rate_schedule = decays 1e-4 to 1e-6 over 35M steps, factor 0.995
    Optimization schedule chosen by hand.
assumptions (5)
  • domain assumption LAMMPS trajectories with the stated harmonic bond and angle potentials, Langevin thermostat, and Parrinello-Rahman barostat are the ground truth for the physics of DEN compression.
    All training and test labels are derived from these simulations; no experimental validation is provided.
  • ad hoc to paper The custom displacement-based estimator nu_est = -Dy/Dx approximates the true Poisson's ratio within 5% for all tested networks.
    Introduced in Supplementary Sec. I.A.6; used for both ground truth and predicted values in parity plots, so evaluation accuracy is bounded by this estimator.
  • domain assumption The message-passing GNN with node velocity and edge vector inputs is sufficiently expressive to learn DEN compression dynamics from few examples.
    This is the core learning assumption; the paper does not prove expressiveness, only demonstrates it empirically on the tested data.
  • domain assumption Forward Euler integration with Delta_t=1 in normalized units is stable enough for the learned acceleration over long rollouts.
    The simulator outputs accelerations that are integrated autoregressively with a fixed coarse time step; stability is assumed and tested only on the reported systems.
  • domain assumption Low-Poisson networks generated by gradient-descent node optimization are representative of the broader class of disordered elastic networks.
    OOD generalization is evaluated on these optimized networks; if they have atypical structure, the extrapolation claim may not transfer to other DEN families.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dynamical Data for More Efficient and Generalizable Learning: A Case Study in Disordered Elastic Networks." pith.science (2026). https://pith.science/paper/GJFOML5M

@misc{pith2026250521125,
  author       = {Pith},
  title        = {Pith review of: Dynamical Data for More Efficient and Generalizable Learning: A Case Study in Disordered Elastic Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GJFOML5M}},
  note         = {Machine review of arXiv:2505.21125}
}
read the original abstract

Machine learning models often require large datasets and struggle to generalize beyond their training distribution. These limitations pose significant challenges in scientific and engineering contexts, where generating exhaustive datasets is often impractical and the goal is frequently to discover novel solutions outside the training domain. In this work, we explore the use of dynamical data through a graph neural network-based simulator to enable efficient system-to-property learning and out-of-distribution prediction in the context of uniaxial compression of two-dimensional disordered elastic networks. We find that the simulator can learn the underlying physical dynamics from a small number of training examples and accurately reproduce the temporal evolution of unseen networks. Notably, the simulator is able to accurately predict emergent properties such as the Poisson's ratio and its dependence on strain, even though it was not explicitly trained for this task. In addition, it generalizes well across variations in system temperature, strain amplitude, and most significantly, Poisson's ratios beyond the training range. These findings suggest that using dynamical data to train machine learning models can support more data efficient and generalizable approaches for materials and molecular design, especially in data-scarce settings.

Figures

Figures reproduced from arXiv: 2505.21125 by the authors.

Figure 1
Figure 1. FIG. 1. Left: Illustration of a two-dimensional disordered elastic [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Predicted network configurations (red) from a compression rollout overlaid on top of the ground truth ones (blue) at specified timesteps [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Predicted and ground truth Poisson’s ratio [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: FIG. 5. Average Poisson’s ratio prediction accuracy after 20 step [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 4
Figure 4. Figure 4: FIG. 4. Individual node trajectory predictions (left) versus ground [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Performance comparison of a model trained with high tem [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Absolute error of Poisson’s ratio prediction for a model [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 1
Figure 1. Figure 1: FIG. 1. Node displacement in [PITH_FULL_IMAGE:figures/full_fig_p014_1.png]
Figure 2
Figure 2. Figure 2: FIG. 2. Behavior of [PITH_FULL_IMAGE:figures/full_fig_p015_2.png]
Figure 3
Figure 3. Figure 3: FIG. 3. Node trajectories for a model trained and tested on data simulated at [PITH_FULL_IMAGE:figures/full_fig_p016_3.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

66 extracted references · 21 canonical work pages

  1. [1]

    Mowbray , author T

    author author M. Mowbray , author T. Savage , author C. Wu , author Z. Song , author B. A. \ Cho , author E. A. \ Del Rio-Chanona , \ and\ author D. Zhang ,\ title title Machine learning for biochemical engineering: A review , \ 10.1016/j.bej.2021.108054 journal journal Biochemical Engineering Journal \ volume 172 ,\ pages 1369--703 ( year 2021 ) NoStop

    work page arXiv

  2. [2]

    Liu , author Z

    author author H. Liu , author Z. Fu , author K. Yang , author X. Xu , \ and\ author M. Bauchy ,\ title title Machine learning for glass science and engineering: A review , \ 10.1016/j.jnoncrysol.2019.04.039 journal journal Journal of Non-Crystalline Solids \ volume 557 ,\ pages 119419 ( year 2021 ) NoStop

    work page doi:10.1016/j.jnoncrysol.2019.04.039 2019

  3. [3]

    Wang , author T

    author author H. Wang , author T. Fu , author Y. Du , author W. Gao , author K. Huang , author Z. Liu , author P. Chandak , author S. Liu , author P. Van Katwyk , author A. Deac , author A. Anandkumar , author K. Bergen , author C. P. \ Gomes , author S. Ho , author P. Kohli , author J. Lasenby , author J. Leskovec , author T. Y. \ Liu , author A. Manrai ...

  4. [4]

    Notin , author N

    author author P. Notin , author N. Rollins , author Y. Gal , author C. Sander , \ and\ author D. Marks ,\ title title Machine learning for functional protein design , \ 10.1038/s41587-024-02127-0 journal journal Nature Biotechnology \ volume 42 ,\ pages 216--228 ( year 2024 ) NoStop

    work page doi:10.1038/s41587-024-02127-0

  5. [5]

    Jumper , author R

    author author J. Jumper , author R. Evans , author A. Pritzel , author T. Green , author M. Figurnov , author O. Ronneberger , author K. Tunyasuvunakool , author R. Bates , author A. Z \' i dek , author A. Potapenko , author A. Bridgland , author C. Meyer , author S. A. \ Kohl , author A. J. \ Ballard , author A. Cowie , author B. Romera-Paredes , author ...

  6. [6]

    Krishna , author J

    author author R. Krishna , author J. Wang , author W. Ahern , author P. Sturmfels , author P. Venkatesh , author I. Kalvet , author G. R. \ Lee , author F. S. \ Morey-Burrows , author I. Anishchenko , author I. R. \ Humphreys , author R. McHugh , author D. Vafeados , author X. Li , author G. A. \ Sutherland , author A. Hitchcock , author C. Neil Hunter , ...

  7. [7]

    Behler \ and\ author M

    author author J. Behler \ and\ author M. Parrinello ,\ title title Generalized neural-network representation of high-dimensional potential-energy surfaces , \ 10.1103/PhysRevLett.98.146401 journal journal Physical Review Letters \ volume 98 ( year 2007 ),\ 10.1103/PhysRevLett.98.146401 NoStop

    doi:10.1103/physrevlett.98.146401

  8. [8]

    author author P. M. \ Piaggi , author A. Selloni , author A. Z. \ Panagiotopoulos , author R. Car , \ and\ author P. G. \ Debenedetti ,\ title title A first-principles machine-learning force field for heterogeneous ice nucleation on microcline feldspar , \ 10.1039/D3FD00100H journal journal Faraday Discussions \ volume 249 ,\ pages 98--113 ( year 2024 ) NoStop

    work page doi:10.1039/d3fd00100h

Show all 66 references
  1. [9]

    Zhang , author H

    author author L. Zhang , author H. Wang , author M. C. \ Muniz , author A. Z. \ Panagiotopoulos , author R. Car , \ and\ author E. Weinan ,\ title title A deep potential model with long-range electrostatic interactions , \ 10.1063/5.0083669 journal journal Journal of Chemical ...

    doi

  2. [10]

    author author A. P. \ Bart \' o k , author J. Kermode , author N. Bernstein , \ and\ author G. Cs \' a nyi ,\ title title Machine Learning a General-Purpose Interatomic Potential for Silicon , \ 10.1103/PhysRevX.8.041048 journal journal Physical Review X \ volume 8 ,\ pages 04...

    doi

  3. [11]

    Cheng , author G

    author author B. Cheng , author G. Mazzola , author C. J. \ Pickard , \ and\ author M. Ceriotti ,\ title title Evidence for supercritical behaviour of high-pressure liquid hydrogen , \ 10.1038/s41586-020-2677-y journal journal Nature \ volume 585 ,\ pages 217--220 ( year 2020 ) NoStop

    doi

  4. [12]

    Zhang , author L

    author author J. Zhang , author L. Bonati , author E. Trizio , author O. Zhang , author Y. Kang , author T. Hou , \ and\ author M. Parrinello ,\ title title Descriptor-Free Collective Variables from Geometric Graph Neural Networks , \ 10.1021/acs.jctc.4c01197 journal journal J...

    doi

  5. [13]

    Bonati , author G

    author author L. Bonati , author G. M. \ Piccini , \ and\ author M. Parrinello ,\ title title Deep learning the slow modes for rare events sampling , \ 10.1073/pnas.2113533118 journal journal Proceedings of the National Academy of Sciences of the United States of America \ vol...

    doi

  6. [14]

    Bonati , author E

    author author L. Bonati , author E. Trizio , author A. Rizzi , \ and\ author M. Parrinello ,\ title title A unified framework for machine learning collective variables for enhanced sampling simulations: mlcolvar , \ 10.1063/5.0156343 journal journal Journal of Chemical Physics...

    doi

  7. [15]

    Mendels , author G

    author author D. Mendels , author G. Piccini , author Z. F. \ Brotzakis , author Y. I. \ Yang , \ and\ author M. Parrinello ,\ title title Folding a small protein using harmonic linear discriminant analysis , \ 10.1063/1.5053566 journal journal Journal of Chemical Physics \ vo...

    doi

  8. [16]

    Mendels , author G

    author author D. Mendels , author G. Piccini , \ and\ author M. Parrinello ,\ title title Collective Variables from Local Fluctuations , \ 10.1021/acs.jpclett.8b00733 journal journal Journal of Physical Chemistry Letters \ volume 9 ,\ pages 2776--2781 ( year 2018 ) NoStop

    doi

  9. [17]

    Piccini , author D

    author author G. Piccini , author D. Mendels , \ and\ author M. Parrinello ,\ title title Metadynamics with Discriminants: A Tool for Understanding Chemistry , \ 10.1021/acs.jctc.8b00634 journal journal Journal of Chemical Theory and Computation \ volume 14 ,\ pages 5040--5044...

    doi

  10. [18]

    Rizzi , author D

    author author V. Rizzi , author D. Mendels , author E. Sicilia , \ and\ author M. Parrinello ,\ title title Blind Search for Complex Chemical Pathways Using Harmonic Linear Discriminant Analysis , \ 10.1021/acs.jctc.9b00358 journal journal Journal of Chemical Theory and Comput...

    doi

  11. [19]

    No \' e , author S

    author author F. No \' e , author S. Olsson , author J. K \" o hler , \ and\ author H. Wu ,\ title title Boltzmann generators: Sampling equilibrium states of many-body systems with deep learning , \ 10.1126/science.aaw1147 journal journal Science \ volume 365 ( year 2019 ),\ 1...

    doi

  12. [20]

    author author S. A. \ Hollingsworth \ and\ author R. O. \ Dror ,\ title title Molecular Dynamics Simulation for All , \ 10.1016/j.neuron.2018.08.011 journal journal Neuron \ volume 99 ,\ pages 1129--1143 ( year 2018 ) NoStop

    2018 doi

  13. [21]

    author author S. S. \ Chong , author Y. S. \ Ng , author H. Q. \ Wang , \ and\ author J. C. \ Zheng ,\ title title Advances of machine learning in materials science: Ideas and techniques , \ 10.1007/s11467-023-1325-z journal journal Frontiers of Physics \ volume 19 ,\ pages 13...

    doi

  14. [22]

    Zhang , author L

    author author X. Zhang , author L. Wang , author J. Helwig , author Y. Luo , author C. Fu , author Y. Xie , author M. Liu , author Y. Lin , author Z. Xu , author K. Yan , author K. Adams , author M. Weiler , author X. Li , author T. Fu , author Y. Wang , author A. Strasser , a...

    2023 arXiv

  15. [23]

    Avery , author X

    author author P. Avery , author X. Wang , author C. Oses , author E. Gossett , author D. M. \ Proserpio , author C. Toher , author S. Curtarolo , \ and\ author E. Zurek ,\ title title Predicting superhard materials via a machine learning informed evolutionary structure search ...

    doi

  16. [24]

    Ojih , author A

    author author J. Ojih , author A. Rodriguez , author J. Hu , \ and\ author M. Hu ,\ title title Screening outstanding mechanical properties and low lattice thermal conductivity using global attention graph neural network , \ 10.1016/j.egyai.2023.100286 journal journal Energy a...

  17. [25]

    Yao , author B

    author author Z. Yao , author B. S \' a nchez-Lengeling , author N. S. \ Bobbitt , author B. J. \ Bucior , author S. G. H. \ Kumar , author S. P. \ Collins , author T. Burns , author T. K. \ Woo , author O. K. \ Farha , author R. Q. \ Snurr , \ and\ author A. Aspuru-Guzik ,\ t...

    doi

  18. [26]

    Liu , author D

    author author Z. Liu , author D. Zhu , author S. P. \ Rodrigues , author K. T. \ Lee , \ and\ author W. Cai ,\ title title Generative Model for the Inverse Design of Metasurfaces , \ 10.1021/acs.nanolett.8b03171 journal journal Nano Letters \ volume 18 ,\ pages 6570--6576 ( ye...

    doi

  19. [27]

    author author J. H. \ Bastek \ and\ author D. M. \ Kochmann ,\ title title Inverse design of nonlinear mechanical metamaterials via video denoising diffusion models , \ 10.1038/s42256-023-00762-x journal journal Nature Machine Intelligence \ volume 5 ,\ pages 1466--1475 ( year...

    doi

  20. [28]

    Mendels \ and\ author J

    author author D. Mendels \ and\ author J. J. \ De Pablo ,\ title title Collective Variables for Free Energy Surface Tailoring: Understanding and Modifying Functionality in Systems Dominated by Rare Events , \ 10.1021/acs.jpclett.2c00317 journal journal Journal of Physical Chem...

    doi

  21. [29]

    Alzubaidi , author J

    author author L. Alzubaidi , author J. Bai , author A. Al-Sabaawi , author J. Santamar \' i a , author A. S. \ Albahri , author B. S. N. \ Al-dabbagh , author M. A. \ Fadhel , author M. Manoufali , author J. Zhang , author A. H. \ Al-Timemy , author Y. Duan , author A. Abdulla...

  22. [30]

    Chang , author Y.-X

    author author R. Chang , author Y.-X. \ Wang , \ and\ author E. Ertekin ,\ title title Towards overcoming data scarcity in materials science: unifying models and datasets with a mixture of experts framework , \ 10.1038/s41524-022-00929-x journal journal npj Computational Mater...

    doi

  23. [31]

    Meredig , author E

    author author B. Meredig , author E. Antono , author C. Church , author M. Hutchinson , author J. Ling , author S. Paradiso , author B. Blaiszik , author I. Foster , author B. Gibbons , author J. Hattrick-Simpers , author A. Mehta , \ and\ author L. Ward ,\ title title Can mac...

    doi

  24. [32]

    Li , author B

    author author K. Li , author B. DeCost , author K. Choudhary , author M. Greenwood , \ and\ author J. Hattrick-Simpers ,\ title title A critical examination of robustness and generalizability of machine learning prediction of materials properties , \ 10.1038/s41524-023-01012-9...

    doi

  25. [33]

    author author S. S. \ Omee , author N. Fu , author R. Dong , author M. Hu , \ and\ author J. Hu ,\ title title Structure-based out-of-distribution (OOD) materials property prediction: a benchmark study , \ 10.1038/s41524-024-01316-4 journal journal npj Computational Materials ...

    doi

  26. [34]

    author author S. K. \ Kauwe , author J. Graser , author R. Murdock , \ and\ author T. D. \ Sparks ,\ title title Can machine learning find extraordinary materials? \ 10.1016/j.commatsci.2019.109498 journal journal Computational Materials Science \ volume 174 ,\ pages 109498 ( ...

    2019

  27. [35]

    Colen , author M

    author author J. Colen , author M. Han , author R. Zhang , author S. A. \ Redford , author L. M. \ Lemma , author L. Morgan , author P. V. \ Ruijgrok , author R. Adkins , author Z. Bryant , author Z. Dogic , author M. L. \ Gardel , author J. J. \ de Pablo , \ and\ author V. Vi...

    doi

  28. [36]

    Cranmer , author A

    author author M. Cranmer , author A. Sanchez-Gonzalez , author P. Battaglia , author R. Xu , author K. Cranmer , author D. Spergel , \ and\ author S. Ho ,\ title title Discovering symbolic models from deep learning with inductive biases , \ http://arxiv.org/abs/2006.11287 http...

    2006 arXiv

  29. [37]

    Wiewel , author M

    author author S. Wiewel , author M. Becher , \ and\ author N. Thuerey ,\ title title Latent Space Physics: Towards Learning the Temporal Evolution of Fluid Flow , \ 10.1111/cgf.13620 journal journal Computer Graphics Forum \ volume 38 ,\ pages 71--82 ( year 2019 ) NoStop

    doi

  30. [38]

    Ladick \' y , author S

    author author L. Ladick \' y , author S. Jeong , author B. Solenthaler , author M. Pollefeys , \ and\ author M. Gross ,\ title title Data-driven Fluid Simulations using Regression Forests , \ 10.1145/2816795.2818129 journal journal ACM Transactions on Graphics \ volume 34 ( ye...

  31. [39]

    Kochkov , author J

    author author D. Kochkov , author J. A. \ Smith , author A. Alieva , author Q. Wang , author M. P. \ Brenner , \ and\ author S. Hoyer ,\ title title Machine learning–accelerated computational fluid dynamics , \ 10.1073/pnas.2101784118 journal journal Proceedings of the Nationa...

    doi

  32. [40]

    Sanchez-Gonzalez , author J

    author author A. Sanchez-Gonzalez , author J. Godwin , author T. Pfaff , author R. Ying , author J. Leskovec , \ and\ author P. W. \ Battaglia ,\ title title Learning to simulate complex physics with graph networks , \ http://arxiv.org/abs/2002.09405 journal journal 37th Inter...

    2002 arXiv

  33. [41]

    Pfaff , author M

    author author T. Pfaff , author M. Fortunato , author A. Sanchez-Gonzalez , \ and\ author P. W. \ Battaglia ,\ title title Learning Mesh-Based Simulation with Graph Networks , \ http://arxiv.org/abs/2010.03409 journal journal ICLR 2021 - 9th International Conference on Learnin...

    2010 arXiv

  34. [42]

    Liu , author Z

    author author H. Liu , author Z. Huang , author S. S. \ Schoenholz , author E. D. \ Cubuk , author M. M. \ Smedskjaer , author Y. Sun , author W. Wang , \ and\ author M. Bauchy ,\ title title Learning molecular dynamics: predicting the dynamics of glasses by a machine learning...

    doi

  35. [43]

    Kumar \ and\ author J

    author author K. Kumar \ and\ author J. Vantassel ,\ title title GNS: A generalizable Graph Neural Network-based simulator for particulate and fluid modeling , \ 10.21105/joss.05025 journal journal Journal of Open Source Software \ volume 8 ,\ pages 5025 ( year 2023 ) NoStop

    doi

  36. [44]

    author author K. R. \ Allen , author T. Lopez-Guevara , author K. Stachenfeld , author A. Sanchez-Gonzalez , author P. Battaglia , author J. Hamrick , \ and\ author T. Pfaff ,\ title title Physical Design using Differentiable Learned Simulators , \ http://arxiv.org/abs/2202.00...

    2022 arXiv

  37. [45]

    Li , author K

    author author Z. Li , author K. Meidani , author P. Yadav , \ and\ author A. Barati Farimani ,\ title title Graph neural networks accelerated molecular dynamics , \ 10.1063/5.0083060 journal journal Journal of Chemical Physics \ volume 156 ( year 2022 ),\ 10.1063/5.0083060 NoStop

    doi

  38. [46]

    Klein , author A

    author author L. Klein , author A. Y. K. \ Foong , author T. E. \ Fjelde , author B. Mlodozeniec , author M. Brockschmidt , author S. Nowozin , author F. No \' e , \ and\ author R. Tomioka ,\ title title Timewarp: Transferable Acceleration of Molecular Dynamics by Learning Tim...

    2023 arXiv

  39. [47]

    Lerner , author E

    author author E. Lerner , author E. Degiuli , author G. D \" u ring , \ and\ author M. Wyart ,\ title title Breakdown of continuum elasticity in amorphous solids , \ 10.1039/c4sm00311j journal journal Soft Matter \ volume 10 ,\ pages 5085--5092 ( year 2014 ) NoStop

    doi

  40. [48]

    author author D. R. \ Reid , author N. Pashine , author J. M. \ Wozniak , author H. M. \ Jaeger , author A. J. \ Liu , author S. R. \ Nagel , \ and\ author J. J. \ De Pablo ,\ title title Auxetic metamaterials from disordered networks , \ 10.1073/pnas.1717442115 journal journa...

    doi

  41. [49]

    author author M. L. \ Manning ,\ title title Rigidity in mechanical biological networks , \ 10.1016/j.cub.2024.07.014 journal journal Current biology : CB \ volume 34 ,\ pages R1024--R1030 ( year 2024 ) NoStop

    2024 doi

  42. [50]

    Berneman \ and\ author D

    author author M. Berneman \ and\ author D. Hexner ,\ title title Designing precise dynamical steady states in disordered networks , \ http://arxiv.org/abs/2409.05060 \ ( year 2024 ) NoStop

    2024 arXiv

  43. [51]

    Mendels , author F

    author author D. Mendels , author F. Byl \' e hn , author T. W. \ Sirk , \ and\ author J. J. \ de Pablo ,\ title title Systematic modification of functionality in disordered elastic networks through free energy surface tailoring , \ 10.1126/sciadv.adf7541 journal journal Scien...

    doi

  44. [52]

    Shen , author M

    author author M. Shen , author M. A. \ Reyes-Martinez , author L. A. \ Powell , author M. A. \ Iadicola , author A. Sharma , author F. Byl \' e hn , author N. Pashine , author E. P. \ Chan , author C. L. \ Soles , author H. M. \ Jaeger , \ and\ author J. J. \ de Pablo ,\ title...

    doi

  45. [53]

    Stern , author D

    author author M. Stern , author D. Hexner , author J. W. \ Rocks , \ and\ author A. J. \ Liu ,\ title title Supervised Learning in Physical Networks: From Machine Learning to Learning Machines , \ 10.1103/PhysRevX.11.021045 journal journal Physical Review X \ volume 11 ,\ page...

    doi

  46. [54]

    Stern \ and\ author A

    author author M. Stern \ and\ author A. Murugan ,\ title title Learning Without Neurons in Physical Systems , \ 10.1146/annurev-conmatphys-040821-113439 journal journal Annual Review of Condensed Matter Physics \ volume 14 ,\ pages 417--441 ( year 2023 ) NoStop

    doi

  47. [55]

    Bahar , author T

    author author I. Bahar , author T. R. \ Lezon , author L. W. \ Yang , \ and\ author E. Eyal ,\ title title Global dynamics of proteins: Bridging between structure and function , \ 10.1146/annurev.biophys.093008.131258 journal journal Annual Review of Biophysics \ volume 39 ,\ ...

  48. [56]

    Parrinello \ and\ author A

    author author M. Parrinello \ and\ author A. Rahman ,\ title title Polymorphic transitions in single crystals: A new molecular dynamics method , \ 10.1063/1.328693 journal journal Journal of Applied Physics \ volume 52 ,\ pages 7182--7190 ( year 1981 ) NoStop

    doi

  49. [57]

    Schneider \ and\ author E

    author author T. Schneider \ and\ author E. Stoll ,\ title title Molecular-dynamics study of a three-dimensional one-component model for distortive phase transitions , \ 10.1103/PhysRevB.17.1302 journal journal Physical Review B \ volume 17 ,\ pages 1302--1322 ( year 1978 ) NoStop

    doi

  50. [58]

    author author J. W. \ Rocks , author N. Pashine , author I. Bischofberger , author C. P. \ Goodrich , author A. J. \ Liu , \ and\ author S. R. \ Nagel ,\ title title Designing allostery-inspired response in mechanical networks , \ 10.1073/pnas.1612139114 journal journal Procee...

    doi

  51. [59]

    author author A. J. \ Liu \ and\ author S. R. \ Nagel ,\ title title The jamming transition and the marginally jammed solid , \ 10.1146/annurev-conmatphys-070909-104045 journal journal Annual Review of Condensed Matter Physics \ volume 1 ,\ pages 347--369 ( year 2010 ) NoStop

    doi

  52. [60]

    author author J. B. \ Hamrick , author K. R. \ Allen , author V. Bapst , author T. Zhu , author K. R. \ McKee , author J. B. \ Tenenbaum , \ and\ author P. W. \ Battaglia ,\ title title Relational inductive bias for physical construction in humans and machines , \ http://arxiv...

    2018 arXiv

  53. [61]

    Shi , author Z

    author author Y. Shi , author Z. Huang , author S. Feng , author H. Zhong , author W. Wang , \ and\ author Y. Sun ,\ title title Masked Label Prediction: Unified Message Passing Model for Semi-Supervised Classification , \ 10.24963/ijcai.2021/214 journal journal IJCAI Internat...

    doi

  54. [62]

    Fey \ and\ author J

    author author M. Fey \ and\ author J. E. \ Lenssen ,\ title title Fast Graph Representation Learning with PyTorch Geometric , \ http://arxiv.org/abs/1903.02428 \ ( year 2019 ) NoStop

    1903 arXiv

  55. [63]

    Paszke , author S

    author author A. Paszke , author S. Gross , author F. Massa , author A. Lerer , author J. Bradbury Google , author G. Chanan , author T. Killeen , author Z. Lin , author N. Gimelshein , author L. Antiga , author A. Desmaison , author A. K. \ Xamla , author E. Yang , author Z. ...

    2019

  56. [64]

    author author D. P. \ Kingma \ and\ author J. Ba ,\ title title Adam: A Method for Stochastic Optimization , \ http://arxiv.org/abs/1412.6980 journal journal 3rd International Conference on Learning Representations, ICLR 2015 - Conference Track Proceedings \ ( year 2014 ) NoStop

    2015 arXiv

  57. [65]

    author author A. P. \ Thompson , author H. M. \ Aktulga , author R. Berger , author D. S. \ Bolintineanu , author W. M. \ Brown , author P. S. \ Crozier , author P. J. \ in 't Veld , author A. Kohlmeyer , author S. G. \ Moore , author T. D. \ Nguyen , author R. Shan , author M...

  58. [66]

    Dold \ and\ author D

    author author D. Dold \ and\ author D. Aranguren van Egmond ,\ title title Differentiable graph-structured models for inverse design of lattice materials , \ 10.1016/j.xcrp.2023.101586 journal journal Cell Reports Physical Science \ volume 4 ,\ pages 101586 ( year 2023 ) NoStop

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.