Pith. sign in

REVIEW 5 major objections 6 minor 54 references

Learning Effective Dynamics across Spatio-Temporal Scales of Complex Flows

T0 review · 5 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read Graph-LED learns effective dynamics of complex flows on unstructured meshes, forecasting cylinder and backward-facing step flows at 900x and 100x speedups.

desk verdict Graph-LED is a credible GNN-autoencoder-plus-transformer surrogate for 2D flows on unstructured meshes, but the headline near-wall accuracy claim rests on evidence the paper doesn't show. read the letter →

arxiv 2502.07990 v1 pith:RJ6YAO6C submitted 2025-02-11 cs.LG physics.comp-phphysics.flu-dyn

Han Gao , Sebastian Kaltenbach , Petros Koumoutsakos This is my paper · ORCID
classification cs.LGphysics.comp-phphysics.flu-dyn
keywords graphneuralnetworksreduced-ordermodelingspatio-temporalforecastingfluiddynamicstransformerunstructuredmesheseffectiveNavier-Stokesequations
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 introduces Graph-LED, a two-stage learning architecture that compresses high-dimensional flow states on unstructured meshes into a low-dimensional latent space with a graph neural network autoencoder, then advances that latent state in time with an attention-based transformer. The central claim is that this two-stage design extracts the effective dynamics of flows that span multiple spatio-temporal scales, and the supporting evidence is accurate forecasting of 2D incompressible flow past a cylinder at Reynolds number 696 and over a backward-facing step at Reynolds number 5000. The paper reports test-time speedups of about 900 times for the cylinder and 100 times for the step relative to the OpenFOAM solver, while still resolving near-wall small-scale features and the wake. A sympathetic reader would care because the method suggests a path to fast CFD surrogates on complex geometries without projecting the flow onto a uniform grid.

What carries the argument

The central object is Graph-LED, a two-part architecture. The spatial part is a mesh-based GNN autoencoder built from the message-passing layer of Eq. (7), where each node aggregates mean-updated edge features from its neighbors through MLPs with residual connections and layer normalization, and nearest-neighbor interpolation transfers field values between fine mesh coordinates and a coarse latent coordinate set. The temporal part is a multi-head attention transformer following the GPT-2 formulation, operating in the low-dimensional latent space with a sliding window of $N_{\mathrm{sw}}$ past states so the rollout is autoregressive and trained in the same autoregressive mode rather than by teacher forcing. The load-bearing design choice is the decoupling: the spatial models are fit first to reconstruct snapshots, and the temporal model is then fit to the latent trajectories those spatial models produce.

What would settle it

Run the same training procedure on the cylinder case with context lengths $N_{\mathrm{sw}} = 4, 8, 16, 32$, and $64$ and compare forecast error at $t=50$ and $t=100$; if the error keeps decreasing as the window grows beyond 32, the claim that a finite-memory latent model captures the effective dynamics is falsified.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that a GNN-based encoder-decoder can reduce an irregular finite-volume mesh to a compact latent state (from 27,127 to 1,024 nodes for the cylinder and from 20,480 to 2,048 for the backward-facing step) without losing the information needed to evolve the flow, and that a GPT-2-style transformer can then autoregressively advance that latent state over long horizons with large time steps. The authors report that decoded forecasts match OpenFOAM in vorticity, pressure, wall shear stress, and lift and drag coefficients, and that the model reproduces the mean and variance of vorticity for the turbulent backward-facing step. They attribute the stability of the rollout to the decoupled design: the encoder and decoder are trained first to reconstruct snapshots, and the transformer is trained afterward on the resulting latent trajectories in a truly autoregressive manner, so no noise injection is needed.

Load-bearing premise

The forecast rests on a fixed-size sliding window: the transformer assumes the flow's effective dynamics are determined by the past $N_{\mathrm{sw}}=32$ latent states for the cylinder and $N_{\mathrm{sw}}=8$ for the step, and the paper does not test whether longer context changes the predictions.

Editorial extensions

If this is right

  • Forecasts for the two test cases run at roughly 900 times (cylinder) and 100 times (backward-facing step) the speed of the OpenFOAM solver.
  • Near-cylinder quantities such as pressure gradient, wall shear stress, and lift and drag coefficients are reported to match OpenFOAM, indicating that fine-scale information survives the latent bottleneck.
  • The model predicts not only instantaneous fields but also the mean and variance of vorticity for the turbulent backward-facing step over one flow-through time.
  • Because the encoder and decoder are graph-based and interpolation-based, the framework applies to variable-size and non-uniform meshes without requiring a uniform grid.
  • The autoregressive transformer requires no noise injection for stable training, unlike the GNN temporal baselines compared in the paper.

Reading between the lines

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

  • If the latent space genuinely carries the effective dynamics, the temporal transformer might transfer across nearby Reynolds numbers or mesh resolutions with only the encoder and decoder retrained; the paper does not test this.
  • The sliding-window lengths (32 for the cylinder, 8 for the step) are fixed per problem without a sensitivity study, so a direct sweep of $N_{\mathrm{sw}}$ would reveal whether the latent dynamics are truly finite-memory.
  • The graph formulation suggests a natural extension to three-dimensional and compressible flows, but the paper's evidence is limited to two incompressible two-dimensional test cases.
  • Since the encoder is trained purely on reconstruction, an a-posteriori check relating reconstruction error in near-wall cells to forecast error would test whether the claimed small-scale accuracy comes from the latent dynamics or from the decoder.
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

5 major / 6 minor

Summary. This paper introduces Graph-LED, a reduced-order modeling framework that combines a GNN-based autoencoder for spatial compression of unstructured mesh data with an attention-based transformer that evolves the latent state autoregressively. The authors validate the method on 2D incompressible flow past a cylinder at Re=696 and flow over a backward-facing step at Re=5000, reporting low relative errors for velocity, pressure, vorticity, and TKE, approximately 900x and 100x test-time speedups, and qualitative agreement of vorticity fields, wall quantities, and lift/drag coefficients. The central claim is that Graph-LED robustly forecasts spatio-temporal physics and accurately captures small-scale near-cylinder effects as well as the wake.

Significance. If supported by complete evidence, Graph-LED would be a useful contribution to data-driven reduced-order modeling for unstructured meshes: the architecture decouples expensive spatial compression from cheap latent temporal evolution, handles variable-size meshes, and reports large inference speedups with substantially lower errors than the MeshGraphNet and NNGraphNet baselines. The paper is also commendable for including a comparison table with multiple metrics, a memory estimate, and explicit hyperparameter details. However, the current manuscript provides no code or data, no error bars or repeated seeds, no quantitative error table for the backward-facing step, and no spatial decomposition of reconstruction versus forecast error; these gaps matter because the headline claims concern small-scale accuracy at selected wall locations.

major comments (5)
  1. [§2.3.3, §3.1, Eq. (15)] The autoencoder is trained solely on snapshot reconstruction (Eq. 15) before the transformer is trained (Eq. 16), so the end-to-end RRMSE in Table 1 is bounded below by the decoder error applied to the true latent trajectory, and the paper does not report this lower bound. Because the abstract's claim of 'small-scale effects close to the cylinder' hinges on near-wall resolution, please report per-region reconstruction error (e.g., boundary-layer cells adjacent to the cylinder) and separate the contributions of encoder-decoder and temporal model to the total error.
  2. [§3.2, Figs. 7, 10, 11] The backward-facing step section provides only qualitative vorticity fields and mean/std profiles, with no quantitative error metrics analogous to Table 1. The paper states that Graph-LED shows 'reasonably good performance' over one flow-through time, but no numbers or thresholds are given; please add quantitative errors for velocity, pressure, and vorticity over the test rollout and explicitly define the forecast horizon in flow-through times.
  3. [Table 1, §3.1, §3.2] All reported errors are single point estimates from one test initialization; no repeated seeds or multiple test initial conditions are shown. The empirical claim of robustness requires at least a small ensemble with mean and spread; please include error bars or per-seed tables and state how many test initial conditions are used.
  4. [Eq. (14), Eq. (16), Eq. (17)] There is an inconsistency in the definition of the temporal model: Eq. (14) defines Fr:(Z0,wFr) -> {Z1,...,ZNt} with no parameter input, while Eq. (16) trains Fr(µi, Z*i,0, wFr) and Eq. (17) evaluates Fr(µi, Zi,0, wFr). The phrase 'keep the parameter-dependent vector' in Eq. (14) suggests conditioning on µ is intended, but the sliding-window sets S_j do not show it; please clarify how µ enters the transformer and make the notation consistent.
  5. [§E, Eq. (14)] The finite-memory assumption in the latent space is a structural modeling choice (Nsw=32 for the cylinder, Nsw=8 for the backward-facing step) that is never tested. Since the transformer's window length determines which temporal dependencies can be captured, a sensitivity study with different Nsw, or at least a diagnostic comparing performance against window length, is needed to substantiate the claim that the learned latent dynamics are effective.
minor comments (6)
  1. [Eq. (15)] The Encoder call lists the coordinate set X_i^1 twice; it should presumably be Encoder(G*, X_i^1, X_i^2, wEncoder) to match Eq. (9).
  2. [Appendix B] The nearest-neighbor interpolation method is attributed to reference [41] in Appendix B, while Section 2.3.2 cites reference [46]; please correct the citation.
  3. [Fig. 8 caption] The caption defines τθ = µ(∂(v·n)/∂n)|θ but does not define n or θ; please add definitions of the normal vector and the angular coordinate.
  4. [§3.1] The first paragraph contains the typo 'accuractly' (should be 'accurately').
  5. [§3.1] The sentence 'Training (4500 snapshots) and testing (500 snapshots)' is ambiguous; please clarify whether these are time steps of a single trajectory or separate trajectories and initial conditions.
  6. [Table 1] The 'Memory per State' entry of 2MB lacks context; please specify the mesh size, the latent dimension, and how the memory is measured.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Graph-LED is an empirical forecasting pipeline whose reported errors are measured against held-out OpenFOAM snapshots, and no prediction reduces by construction to a fitted value.

full rationale

Graph-LED is an empirical surrogate-modeling paper, not a derivation from first principles, so the circularity tests apply mainly to whether any claimed prediction is forced by definition, fitting, or self-citation. The spatial autoencoder is trained with the reconstruction loss in Eq. (15) on training snapshots, and the temporal transformer is trained with the latent-space loss in Eq. (16) on encoded training snapshots. Forecasts are then evaluated on held-out OpenFOAM data: Table 1 reports RRMSE, FID, and CRPS over 100 prediction steps for the cylinder, and Figures 5, 7, 8, 9, 10, and 11 compare vorticity, wall quantities, and statistical moments against OpenFOAM. None of these quantities is a refit of the reported test outputs; the test data are not used in either loss function. The abstract's claim about near-cylinder small scales is supported qualitatively by Figures 8 and 9, and although the paper does not separate reconstruction error from forecast error, that is a support-strength or evaluation-completeness concern, not a circularity concern. The self-citations present, such as Refs. [33] and [49], are used as related-work context or as a comparison baseline, not as load-bearing evidence for the current results. There is no imported uniqueness theorem, no ansatz smuggled in via citation, and no renaming of a known empirical pattern as a new organization. The conclusion's acknowledgment of error propagation as a limitation is consistent with the empirical framing and does not indicate that any result is equivalent to its input by construction.

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

The central claim rests on hand-chosen latent dimensions and context windows, on the adequacy of OpenFOAM data as ground truth, and on the untested assumption that a reconstruction-trained autoencoder yields a latent space in which a transformer can evolve the dynamics. No new physical entity is introduced.

free parameters (4)
  • Latent dimensionality (cylinder) = 1024 points
    This resolution is chosen by hand, not by an automated selection, and it determines whether near-wall small scales survive compression.
  • Latent dimensionality (backward-facing step) = 2048 points
    This resolution is chosen by hand for the second test case and is not justified by an information criterion.
  • Transformer context length = 32 for cylinder, 8 for backward-facing step
    The sliding-window length sets the assumed memory of the latent dynamics; no sensitivity study is provided.
  • GNN depth and hidden width = 3 or 5 GNN layers, 128 hidden units
    These architecture choices are not ablated, so their effect on the central accuracy claim is unknown.
assumptions (5)
  • domain assumption Navier-Stokes equations with constant density and viscosity are the governing model (Eqs. 1 and 2).
    The training data are finite-volume solutions of these equations, so the surrogate inherits all modeling error from that choice.
  • domain assumption OpenFOAM solutions are treated as ground truth (Section 3.1, Section 3.2).
    All reported errors compare Graph-LED to OpenFOAM, not to experimental data.
  • ad hoc to paper A GNN autoencoder trained purely on reconstruction loss preserves dynamics-relevant information (Section 2.3.3 and Appendix C).
    The paper does not establish a relationship between reconstruction accuracy and forecast accuracy in latent space.
  • ad hoc to paper Latent dynamics is a finite-memory Markov process with window length Nsw (Eq. 14 and Appendix E).
    The transformer is restricted to a sliding window, yet no test verifies that the true latent process has this memory.
  • ad hoc to paper Nearest-neighbor interpolation is sufficient for coordinate reduction and recovery (Section 2.3.2 and Appendix B).
    This interpolation choice is not compared with alternatives, and it directly affects which information is preserved during compression.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Effective Dynamics across Spatio-Temporal Scales of Complex Flows." pith.science (2026). https://pith.science/paper/RJ6YAO6C

@misc{pith2026250207990,
  author       = {Pith},
  title        = {Pith review of: Learning Effective Dynamics across Spatio-Temporal Scales of Complex Flows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RJ6YAO6C}},
  note         = {Machine review of arXiv:2502.07990}
}
read the original abstract

Modeling and simulation of complex fluid flows with dynamics that span multiple spatio-temporal scales is a fundamental challenge in many scientific and engineering domains. Full-scale resolving simulations for systems such as highly turbulent flows are not feasible in the foreseeable future, and reduced-order models must capture dynamics that involve interactions across scales. In the present work, we propose a novel framework, Graph-based Learning of Effective Dynamics (Graph-LED), that leverages graph neural networks (GNNs), as well as an attention-based autoregressive model, to extract the effective dynamics from a small amount of simulation data. GNNs represent flow fields on unstructured meshes as graphs and effectively handle complex geometries and non-uniform grids. The proposed method combines a GNN based, dimensionality reduction for variable-size unstructured meshes with an autoregressive temporal attention model that can learn temporal dependencies automatically. We evaluated the proposed approach on a suite of fluid dynamics problems, including flow past a cylinder and flow over a backward-facing step over a range of Reynolds numbers. The results demonstrate robust and effective forecasting of spatio-temporal physics; in the case of the flow past a cylinder, both small-scale effects that occur close to the cylinder as well as its wake are accurately captured.

Figures

Figures reproduced from arXiv: 2502.07990 by the authors.

Figure 1
Figure 1. Overview of Graph-LED: A high-dimensional initial state [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Left: Visualization of a FV mesh that can be naturally represented by a Graph. Right: [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. The system is reduced from 27,127 points to 1,024 points. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: Vorticity ( ∂v ∂x − ∂u ∂y ) forecasted by Graph-LED (left), OpenFOAM (middle) and the error (right) from t = 0, 50, 100 (from top row to bottom row). Given a testing initial condition, the predictions using the proposed Graph-LED and OpenFOAM are plotted in [PITH_FULL…
Figure 6
Figure 6. Figure 6: The system is reduced from 20,480 points to 2,048 points. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Vorticity ( ∂v ∂x − ∂u ∂y ) forecasted by Graph-LED (left), OpenFOAM (middle) and the error (right) from t = 0, 0.5, 1, 1.5, 2 (from top row to bottom row). As a second case we consider the turbulent flow over a backward-facing step at Re = 5000, a clas￾sical problem i…
Figure 8
Figure 8. Figure 8: Pressure gradient and wall shear stress ( [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: Lift and drag coefficients of the cylinder forecasted over time by Graph-LED and Open [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: Mean of vorticity over time at different vertical lines over the domain predicted by Open [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Standard deviation (std) of vorticity over time at different vertical lines over the domain [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 46 canonical work pages

  1. [1]

    Multiscale model for turbulent flows.AIAA journal, 26(11):1311–1320, 1988

    David C Wilcox. Multiscale model for turbulent flows.AIAA journal, 26(11):1311–1320, 1988

    work page 1988

  2. [2]

    Exposure science in the 21st century: a vision and a strategy

    National Research Council, Division on Earth, Life Studies, Board on Environmental Studies, Committee on Human, and Environmental Exposure Science in the 21st Century. Exposure science in the 21st century: a vision and a strategy. 2012

    work page 2012

  3. [3]

    Theimpactofsubmesoscalephysicsonprimaryproductivityofplankton

    AmalaMahadevan. Theimpactofsubmesoscalephysicsonprimaryproductivityofplankton. Annual review of marine science, 8(1):161–184, 2016

    work page 2016

  4. [4]

    Multiscale modeling in biomechanics and mechanobiology

    Suvranu De, Wonmuk Hwang, and Ellen Kuhl. Multiscale modeling in biomechanics and mechanobiology. Springer, 2015

    work page 2015

  5. [5]

    Multiscale modeling meets machine learning: What can we learn?Archives of Com- putational Methods in Engineering, 28:1017–1037, 2021

    GraceCYPeng,MarkAlber,AdrianBuganzaTepole,WilliamRCannon,SuvranuDe,Savador Dura-Bernal, Krishna Garikipati, George Karniadakis, William W Lytton, Paris Perdikaris, et al. Multiscale modeling meets machine learning: What can we learn?Archives of Com- putational Methods in Engineering, 28:1017–1037, 2021

    work page 2021

  6. [6]

    Theproperorthogonaldecompositioninthe analysis of turbulent flows.Annual review of fluid mechanics, 25(1):539–575, 1993

    GalBerkooz,PhilipHolmes,andJohnLLumley. Theproperorthogonaldecompositioninthe analysis of turbulent flows.Annual review of fluid mechanics, 25(1):539–575, 1993

    work page 1993

  7. [7]

    Data-driven operator inference for nonintrusive projection-based model reduction

    Benjamin Peherstorfer and Karen Willcox. Data-driven operator inference for nonintrusive projection-based model reduction. Computer Methods in Applied Mechanics and Engineering, 306:196–215, 2016

    work page 2016

  8. [8]

    Nonlinear em- beddings for conserving hamiltonians and other quantities with neural galerkin schemes

    Paul Schwerdtner, Philipp Schulze, Jules Berman, and Benjamin Peherstorfer. Nonlinear em- beddings for conserving hamiltonians and other quantities with neural galerkin schemes. SIAM Journal on Scientific Computing, 46(5):C583–C607, 2024

    work page 2024

Show all 54 references
  1. [9]

    Learning mesh-based simulation with graph networks.arXiv preprint arXiv:2010.03409, 2020

    Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter W Battaglia. Learning mesh-based simulation with graph networks.arXiv preprint arXiv:2010.03409, 2020

    2010 arXiv

  2. [10]

    Physical design using differentiable learned simulators

    Kelsey R Allen, Tatiana Lopez-Guevara, Kimberly Stachenfeld, Alvaro Sanchez-Gonzalez, Pe- ter Battaglia, Jessica Hamrick, and Tobias Pfaff. Physical design using differentiable learned simulators. arXiv preprint arXiv:2202.00728, 2022

    2022 arXiv

  3. [11]

    Learning to simulate complex physics with graph networks

    Alvaro Sanchez-Gonzalez, Jonathan Godwin, Tobias Pfaff, Rex Ying, Jure Leskovec, and Peter Battaglia. Learning to simulate complex physics with graph networks. InInternational Confer- ence on Machine Learning, pages 8459–8468. PMLR, 2020

    2020

  4. [12]

    Conditionally parameterized, discretization-awareneuralnetworksformesh-basedmodelingofphysicalsystems

    Jiayang Xu, Aniruddhe Pradhan, and Karthikeyan Duraisamy. Conditionally parameterized, discretization-awareneuralnetworksformesh-basedmodelingofphysicalsystems. Advances in Neural Information Processing Systems, 34:1634–1645, 2021

    2021

  5. [13]

    Graph element networks: adaptive, structured computa- tion and memory

    Ferran Alet, Adarsh Keshav Jeewajee, Maria Bauza Villalonga, Alberto Rodriguez, Tomas Lozano-Perez, and Leslie Kaelbling. Graph element networks: adaptive, structured computa- tion and memory. InInternational Conference on Machine Learning, pages 212–222. PMLR, 2019

    2019

  6. [14]

    End-to-end differentiable physics for learning and control.Advances in neural information pro- cessing systems, 31, 2018

    Filipe de Avila Belbute-Peres, Kevin Smith, Kelsey Allen, Josh Tenenbaum, and J Zico Kolter. End-to-end differentiable physics for learning and control.Advances in neural information pro- cessing systems, 31, 2018. 10

    2018

  7. [15]

    Graph networks as learnable physics engines for inferenceandcontrol

    Alvaro Sanchez-Gonzalez, Nicolas Heess, Jost Tobias Springenberg, Josh Merel, Martin Ried- miller, Raia Hadsell, and Peter Battaglia. Graph networks as learnable physics engines for inferenceandcontrol. In InternationalConferenceonMachineLearning ,pages4470–4479.PMLR, 2018

    2018

  8. [16]

    Modelingthedynamicsofpdesystemswithphysics- constrained deep auto-regressive networks.Journal of Computational Physics, 403:109056, 2020

    NicholasGenevaandNicholasZabaras. Modelingthedynamicsofpdesystemswithphysics- constrained deep auto-regressive networks.Journal of Computational Physics, 403:109056, 2020

    2020

  9. [17]

    Learned coarse models for efficient turbulence simulation.arXiv preprint arXiv:2112.15275, 2021

    Kimberly Stachenfeld, Drummond B Fielding, Dmitrii Kochkov, Miles Cranmer, Tobias Pfaff, Jonathan Godwin, Can Cui, Shirley Ho, Peter Battaglia, and Alvaro Sanchez-Gonzalez. Learned coarse models for efficient turbulence simulation.arXiv preprint arXiv:2112.15275, 2021

    2021 arXiv

  10. [18]

    Professor forcing: A new algorithm for training recurrent net- works

    AlexMLamb,AnirudhGoyalALIASPARTHGOYAL,YingZhang,SaizhengZhang,AaronC Courville, and Yoshua Bengio. Professor forcing: A new algorithm for training recurrent net- works. Advances in neural information processing systems, 29, 2016

    2016

  11. [19]

    Data-driven forecasting of high-dimensional chaotic systems with long short-term memory networks

    PantelisRVlachas,WonminByeon,ZhongYWan,ThemistoklisPSapsis,andPetrosKoumout- sakos. Data-driven forecasting of high-dimensional chaotic systems with long short-term memory networks. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 474(2213):...

    2018

  12. [20]

    Phycrnet: Physics-informed convolutional-recurrentnetworkforsolvingspatiotemporalpdes

    PuRen,ChengpingRao,YangLiu,Jian-XunWang,andHaoSun. Phycrnet: Physics-informed convolutional-recurrentnetworkforsolvingspatiotemporalpdes. ComputerMethodsinApplied Mechanics and Engineering, 389:114399, 2022

    2022

  13. [21]

    Transformers for modeling physical systems.Neural Networks, 146:272–289, 2022

    Nicholas Geneva and Nicholas Zabaras. Transformers for modeling physical systems.Neural Networks, 146:272–289, 2022

    2022

  14. [22]

    Long short-term memory.Neural computation, 9 (8):1735–1780, 1997

    Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory.Neural computation, 9 (8):1735–1780, 1997

    1997

  15. [23]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

    2017

  16. [24]

    Deep learning for universal linear em- beddings of nonlinear dynamics.Nature communications, 9(1):1–10, 2018

    Bethany Lusch, J Nathan Kutz, and Steven L Brunton. Deep learning for universal linear em- beddings of nonlinear dynamics.Nature communications, 9(1):1–10, 2018

    2018

  17. [25]

    A deeplearningenablerfornonintrusivereducedordermodelingoffluidflows

    Suraj Pawar, SM Rahman, H Vaddireddy, Omer San, Adil Rasheed, and Prakash Vedula. A deeplearningenablerfornonintrusivereducedordermodelingoffluidflows. PhysicsofFluids , 31(8):085101, 2019

    2019

  18. [26]

    Modelreductionofdynamicalsystemsonnonlinearman- ifolds using deep convolutional autoencoders.Journal of Computational Physics, 404:108973, 2020

    KookjinLeeandKevinTCarlberg. Modelreductionofdynamicalsystemsonnonlinearman- ifolds using deep convolutional autoencoders.Journal of Computational Physics, 404:108973, 2020

    2020

  19. [27]

    Machine-learning-based reduced-ordermodelingforunsteadyflowsaroundbluffbodiesofvariousshapes

    Kazuto Hasegawa, Kai Fukami, Takaaki Murata, and Koji Fukagata. Machine-learning-based reduced-ordermodelingforunsteadyflowsaroundbluffbodiesofvariousshapes. Theoretical and Computational Fluid Dynamics, 34(4):367–383, 2020

    2020

  20. [28]

    Nonlinearmodedecompositionwithconvo- lutional neural networks for fluid dynamics.Journal of Fluid Mechanics, 882, 2020

    TakaakiMurata,KaiFukami,andKojiFukagata. Nonlinearmodedecompositionwithconvo- lutional neural networks for fluid dynamics.Journal of Fluid Mechanics, 882, 2020

    2020

  21. [29]

    Recurrentneuralnetworksandkoopman-basedframeworksfortemporalpredictionsinalow- order model of turbulence.International Journal of Heat and Fluid Flow, 90:108816, 2021

    Hamidreza Eivazi, Luca Guastoni, Philipp Schlatter, Hossein Azizpour, and Ricardo Vinuesa. Recurrentneuralnetworksandkoopman-basedframeworksfortemporalpredictionsinalow- order model of turbulence.International Journal of Heat and Fluid Flow, 90:108816, 2021. 11

    2021

  22. [30]

    Pierre Jacquier, Azzedine Abdedou, Vincent Delmas, and Azzeddine Soulaïmani. Non- intrusive reduced-order modeling using uncertainty-aware deep neural networks and proper orthogonal decomposition: application to flood modeling.Journal of Computational Physics, 424:109854, 2021

    2021

  23. [31]

    Convolutional neuralnetworksforfluidflowanalysis: towardeffectivemetamodelingandlowdimensional- ization

    Masaki Morimoto, Kai Fukami, Kai Zhang, Aditya G Nair, and Koji Fukagata. Convolutional neuralnetworksforfluidflowanalysis: towardeffectivemetamodelingandlowdimensional- ization. Theoretical and Computational Fluid Dynamics, 35(5):633–658, 2021

    2021

  24. [32]

    Pod-dl-rom: enhancing deep learning-based reduced ordermodelsfornonlinearparametrizedpdesbyproperorthogonaldecomposition

    Stefania Fresca and Andrea Manzoni. Pod-dl-rom: enhancing deep learning-based reduced ordermodelsfornonlinearparametrizedpdesbyproperorthogonaldecomposition. Computer Methods in Applied Mechanics and Engineering, 388:114181, 2022

    2022

  25. [33]

    Mul- tiscale simulations of complex systems by learning their effective dynamics.Nature Machine Intelligence, 4(4):359–366, 2022

    Pantelis R Vlachas, Georgios Arampatzis, Caroline Uhler, and Petros Koumoutsakos. Mul- tiscale simulations of complex systems by learning their effective dynamics.Nature Machine Intelligence, 4(4):359–366, 2022

    2022

  26. [34]

    Data-driven reduced order model with temporal convolutional neural network

    Pin Wu, Junwu Sun, Xuting Chang, Wenjie Zhang, Rossella Arcucci, Yike Guo, and Christo- pher C Pain. Data-driven reduced order model with temporal convolutional neural network. Computer Methods in Applied Mechanics and Engineering, 360:112766, 2020

    2020

  27. [35]

    Deepneu- ral networks for nonlinear model order reduction of unsteady flows.Physics of Fluids, 32(10): 105104, 2020

    HamidrezaEivazi,HadiVeisi,MohammadHosseinNaderi,andVahidEsfahanian. Deepneu- ral networks for nonlinear model order reduction of unsteady flows.Physics of Fluids, 32(10): 105104, 2020

    2020

  28. [36]

    Reduced-order modeling of advection-dominated systems with recurrent neural networks and convolutional autoen- coders

    Romit Maulik, Bethany Lusch, and Prasanna Balaprakash. Reduced-order modeling of advection-dominated systems with recurrent neural networks and convolutional autoen- coders. Physics of Fluids, 33(3):037106, 2021

    2021

  29. [37]

    Unsteady reduced-ordermodelofflowovercylindersbasedonconvolutionalanddeconvolutionalneu- ral network structure.Physics of Fluids, 32(12):123609, 2020

    Jiang-Zhou Peng, Siheng Chen, Nadine Aubry, Zhihua Chen, and Wei-Tao Wu. Unsteady reduced-ordermodelofflowovercylindersbasedonconvolutionalanddeconvolutionalneu- ral network structure.Physics of Fluids, 32(12):123609, 2020

    2020

  30. [38]

    Multi-levelconvolutionalautoencodernetworksforpara- metric prediction of spatio-temporal dynamics.Computer Methods in Applied Mechanics and Engineering, 372:113379, 2020

    JiayangXuandKarthikDuraisamy. Multi-levelconvolutionalautoencodernetworksforpara- metric prediction of spatio-temporal dynamics.Computer Methods in Applied Mechanics and Engineering, 372:113379, 2020

    2020

  31. [39]

    Eigen-gnn: A graph structure preserving plug-in for gnns.IEEE Transactions on Knowledge and Data Engineering, 35(3):2544– 2555, 2021

    Ziwei Zhang, Peng Cui, Jian Pei, Xin Wang, and Wenwu Zhu. Eigen-gnn: A graph structure preserving plug-in for gnns.IEEE Transactions on Knowledge and Data Engineering, 35(3):2544– 2555, 2021

    2021

  32. [40]

    InSC24-W: Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis, pages 1058–1070

    Shivam Barwey, Riccardo Balin, Bethany Lusch, Saumil Patel, Ramesh Balakrishnan, Pinaki Pal,RomitMaulik,andVenkatramVishwanath.Scalableandconsistentgraphneuralnetworks for distributed mesh-based data-driven modeling. InSC24-W: Workshops of the International Conference for High...

    2024

  33. [41]

    Gnumap: A parameter-free approach to unsu- perviseddimensionalityreductionviagraphneuralnetworks

    Jihee You, So Won Jeong, and Claire Donnat. Gnumap: A parameter-free approach to unsu- perviseddimensionalityreductionviagraphneuralnetworks. arXivpreprintarXiv:2407.21236 , 2024

    2024 arXiv

  34. [42]

    Interpretable a-posteriori error indication for graph neural network surrogate models.Computer Methods in Applied Mechanics and Engineer- ing, 433:117509, 2025

    Shivam Barwey, Hojin Kim, and Romit Maulik. Interpretable a-posteriori error indication for graph neural network surrogate models.Computer Methods in Applied Mechanics and Engineer- ing, 433:117509, 2025

    2025

  35. [43]

    The finite volume method in computational fluid dynamics, fluid mechanics and its applications.Cham: Springer, 113, 2016

    F Moukalled, L Mangani, and M Darwish. The finite volume method in computational fluid dynamics, fluid mechanics and its applications.Cham: Springer, 113, 2016. 12

    2016

  36. [44]

    Openfoam: Ac++libraryforcomplex physicssimulations

    HrvojeJasak,AleksandarJemcov,ZeljkoTukovic,etal. Openfoam: Ac++libraryforcomplex physicssimulations. InInternationalworkshoponcoupledmethodsinnumericaldynamics ,volume 1000, pages 1–20. IUC Dubrovnik Croatia, 2007

    2007

  37. [45]

    Layer normalization.arXiv preprint arXiv:1607.06450, 2016

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E Hinton. Layer normalization.arXiv preprint arXiv:1607.06450, 2016

    2016 arXiv

  38. [46]

    Pointnet++: Deephierarchical featurelearningonpointsetsinametricspace

    CharlesRuizhongtaiQi,LiYi,HaoSu,andLeonidasJGuibas. Pointnet++: Deephierarchical featurelearningonpointsetsinametricspace. Advancesinneuralinformationprocessingsystems , 30, 2017

    2017

  39. [47]

    Lan- guage models are unsupervised multitask learners.OpenAI blog, 1(8):9, 2019

    AlecRadford,JeffreyWu,RewonChild,DavidLuan,DarioAmodei,IlyaSutskever,etal. Lan- guage models are unsupervised multitask learners.OpenAI blog, 1(8):9, 2019

    2019

  40. [48]

    Multi-fidelity generative deep learning turbulent flows

    Nicholas Geneva and Nicholas Zabaras. Multi-fidelity generative deep learning turbulent flows. arXiv preprint arXiv:2006.04731, 2020

    2006 arXiv

  41. [49]

    NatureCommunications,15(1):8904,2024

    HanGao,SebastianKaltenbach,andPetrosKoumoutsakos.Generativelearningforforecasting thedynamicsofhigh-dimensionalcomplexsystems. NatureCommunications,15(1):8904,2024

    2024

  42. [50]

    Adam: Amethodforstochasticoptimization

    DiederikPKingmaandJimmyBa. Adam: Amethodforstochasticoptimization. arXivpreprint arXiv:1412.6980, 2014

    2014 arXiv

  43. [51]

    Neural message passing for quantum chemistry

    JustinGilmer,SamuelSSchoenholz,PatrickFRiley,OriolVinyals,andGeorgeEDahl. Neural message passing for quantum chemistry. InInternational conference on machine learning, pages 1263–1272. PMLR, 2017

    2017

  44. [52]

    Turbulentflows

    StephenBPope. Turbulentflows. MeasurementScienceandTechnology ,12(11):2020–2021,2001

    2020

  45. [53]

    Rethinkingfid: Towardsabetterevaluationmetricforimagegeneration

    SadeepJayasumana,SrikumarRamalingam,AndreasVeit,DanielGlasner,AyanChakrabarti, andSanjivKumar. Rethinkingfid: Towardsabetterevaluationmetricforimagegeneration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9307– 9315, 2024

    2024

  46. [54]

    Decomposition of the continuous ranked probability score for ensemble pre- diction systems.Weather and Forecasting, 15(5):559–570, 2000

    Hans Hersbach. Decomposition of the continuous ranked probability score for ensemble pre- diction systems.Weather and Forecasting, 15(5):559–570, 2000. 13 A. GNN Architecture - Training Details The success of Encoder and Decoder is inherently linked to the proper network weigh...

    2000

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.