pith. sign in

arxiv: 2510.00233 · v2 · submitted 2025-09-30 · 💻 cs.LG · physics.flu-dyn

Differentiable Autoencoding Neural Operator for Interpretable and Integrable Latent Space Modeling

Siva Viknesh , Amirhossein Arzani This is my paper

Pith reviewed 2026-05-18 11:22 UTC · model grok-4.3

classification 💻 cs.LG physics.flu-dyn
keywords neural operatorsautoencoderslatent space modelingphysics-informed machine learningspatiotemporal datadifferentiable PDE solversfluid flow simulation
0
0 comments X

The pith

DIANO embeds a differentiable PDE solver in a neural operator autoencoder to evolve low-fidelity physics on a coarse latent grid for accurate high-fidelity reconstruction.

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

The paper introduces DIANO, an autoencoding neural operator that maps high-dimensional spatio-temporal inputs to a coarse-grid latent space using encoding and decoding operators. A fully differentiable PDE solver is placed as the only functional mapping inside this latent space, so end-to-end training occurs with governing equations prescribed in advance through parametric PDEs. The method is evaluated on flow past a cylinder, stenosed arteries, and patient-specific coronary flows, where reconstruction quality tracks the fidelity of the embedded latent PDE relative to the true physics. Sympathetic readers care because the approach simultaneously reduces dimensionality, enforces interpretability, and cuts computational cost by evolving cheap coarse-grid dynamics instead of the full high-resolution system.

Core claim

DIANO constructs visualizable coarse-grid latent spaces for both dimensionality and geometric reduction across varying spatial discretizations, with governing equations enforced directly within the latent space. An encoding neural operator coarsens the high-dimensional input functions into the latent representation while a decoding neural operator reconstructs the original inputs via spatial refinement. A fully differentiable PDE solver is integrated as the sole input-output functional mapping operator within the latent space, enabling end-to-end training with physics prescribed a priori through parametric PDEs such as 2D unsteady advection-diffusion and the 3D Pressure-Poisson equation. The

What carries the argument

The central mechanism is the placement of a fully differentiable PDE solver as the sole input-output operator inside the latent space of an autoencoding neural operator, so that coarse-grid latent evolution follows prescribed parametric physics while encoding and decoding operators handle spatial coarsening and refinement.

If this is right

  • Reconstruction of high-fidelity spatio-temporal fields becomes possible by evolving low-fidelity PDEs on the coarse latent grid at reduced computational cost.
  • Latent representations become coherent, spatially organized, and meaningful when the embedded PDE matches the underlying physics.
  • Accuracy and representation quality are directly governed by the fidelity of the chosen latent PDE relative to the true dynamics.
  • The framework works across different spatial discretizations and PDE types including advection-diffusion and Pressure-Poisson equations.
  • Performance on fluid benchmarks exceeds or matches convolutional neural operators and standard autoencoders while adding explicit physics constraints.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same coarse latent grid could serve as a natural interface for coupling multiple physics models or for multi-fidelity hierarchies in engineering design.
  • If the latent PDE itself were allowed to adapt during training, the method might extend to regimes where the exact governing equations are only partially known.
  • Because the entire pipeline remains differentiable, the latent evolution could be embedded inside gradient-based optimization loops for control or inverse problems.
  • Limits are likely to appear in strongly chaotic or turbulent regimes where low-fidelity approximations on coarse grids break down.

Load-bearing premise

A low-fidelity version of the governing PDE, when solved on the coarse latent grid produced by the encoder, can still capture the essential dynamics of the original high-dimensional system sufficiently well for accurate reconstruction.

What would settle it

Reconstruction error stays high or latent fields fail to organize spatially even after the embedded PDE fidelity is raised to closely match the true physics on the same benchmark flows.

Figures

Figures reproduced from arXiv: 2510.00233 by Amirhossein Arzani, Siva Viknesh.

Figure 1
Figure 1. Figure 1: (a) Schematic of the proposed DIANO framework for spatiotemporal modeling. (b) Fourier Neural Operator [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: DIANO architectural variants for each modeling scenario. The [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Benchmark flow problems considered in this study. (a) Flow past a 2D circular cylinder. (b) Flow through [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Nonlinear dimensionality reduction - Static Mapping. Comparison of vorticity latent space structure [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Nonlinear Dimensionality Reduction with Temporal marching. Comparison of vorticity latent space struc [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Geometrical Reduction with Temporal Marching. The 2D stenosis example is considered and spatio [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Many-to-One Functional Mapping. The DIANO framework infers a single pressure field from three input [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
read the original abstract

Scientific machine learning has enabled the extraction of physical insights and data-driven modeling of high-dimensional spatiotemporal data, yet achieving physically interpretable latent representations and computationally efficient surrogates remains an open challenge. We propose the DIfferentiable Autoencoding Neural Operator - DIANO, an autoencoding neural operator framework that constructs visualizable coarse-grid latent spaces for both dimensionality and geometric reduction across varying spatial discretizations, with governing equations enforced directly within the latent space. Built upon neural operators, DIANO achieves this through an encoding neural operator that spatially coarsens the high-dimensional input functions into the latent representation, and a decoding neural operator that reconstructs the original inputs via spatial refinement. We assess DIANO's latent representation and performance against baselines, including the Convolutional Neural Operator and standard autoencoders. Furthermore, a fully differentiable partial differential equation (PDE) solver is integrated as the sole input-output functional mapping operator within the latent space, enabling end-to-end training with governing physics prescribed a priori through parametric PDEs. Various PDE formulations are investigated, including the 2D unsteady advection-diffusion and the 3D Pressure--Poisson equation, revealing that the fidelity of the embedded PDE relative to the true physics governs the learned latent representation and reconstruction accuracy. Benchmark problems include flow past a 2D cylinder, flow through a 2D symmetric stenosed artery, and a 3D patient-specific coronary artery, showing accurate reconstruction of high-fidelity spatio-temporal fields through low-fidelity latent PDE evolution at reduced computational cost, while yielding coherent, spatially organized, and meaningful latent structures.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes DIANO, a differentiable autoencoding neural operator that encodes high-dimensional spatio-temporal fields into a spatially coarsened latent grid via a neural operator encoder, evolves the latent representation by solving a prescribed low-fidelity parametric PDE (such as 2D unsteady advection-diffusion or 3D Pressure-Poisson) using a fully differentiable solver, and reconstructs the original fields via a neural operator decoder. It evaluates the approach on fluid flow benchmarks including flow past a 2D cylinder, flow through a 2D stenosed artery, and 3D patient-specific coronary artery flow, claiming accurate high-fidelity reconstructions at reduced cost, coherent spatially organized latent structures, and advantages over baselines such as the Convolutional Neural Operator and standard autoencoders, with the fidelity of the embedded PDE directly governing representation quality.

Significance. If the central claims are substantiated with quantitative evidence, the work would offer a meaningful advance in scientific machine learning by enabling end-to-end physics-informed latent space modeling that combines dimensionality reduction, geometric coarsening, and interpretable PDE evolution. The integration of a fully differentiable PDE solver as the sole latent-space operator is a technical strength that allows a priori prescription of governing physics and supports joint optimization.

major comments (2)
  1. Abstract: The central claim of 'accurate reconstruction of high-fidelity spatio-temporal fields through low-fidelity latent PDE evolution' is presented without any quantitative metrics, error bars, ablation studies, or detailed experimental results. This absence directly undermines assessment of whether the low-fidelity latent PDE on the encoder-produced coarse grid captures essential dynamics or whether the decoder compensates for mismatches.
  2. The manuscript states that 'the fidelity of the embedded PDE relative to the true physics governs the learned latent representation and reconstruction accuracy,' yet provides no independent verification (e.g., comparison of latent trajectories to projected high-fidelity solutions) that the jointly trained encoder produces a representation whose coarse-grid evolution under the prescribed PDE remains close to the true projected dynamics rather than being corrected downstream.
minor comments (2)
  1. The description of baseline comparisons (Convolutional Neural Operator and standard autoencoders) would benefit from explicit details on architecture matching, training protocols, and hyperparameter selection to ensure fair evaluation.
  2. Notation for the latent grid resolution and coarsening factor should be introduced consistently with the listed free parameters to improve clarity of the dimensionality reduction mechanism.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback on our manuscript. We appreciate the opportunity to clarify and strengthen the presentation of our results. Below we respond point by point to the major comments and describe the revisions we will make.

read point-by-point responses

  1. Referee: [—] Abstract: The central claim of 'accurate reconstruction of high-fidelity spatio-temporal fields through low-fidelity latent PDE evolution' is presented without any quantitative metrics, error bars, ablation studies, or detailed experimental results. This absence directly undermines assessment of whether the low-fidelity latent PDE on the encoder-produced coarse grid captures essential dynamics or whether the decoder compensates for mismatches.

    Authors: We agree that the abstract would benefit from explicit quantitative support for the central claim. In the revised manuscript we will expand the abstract to include representative quantitative results, such as relative L2 reconstruction errors on the flow-past-cylinder and stenosed-artery benchmarks, together with brief statements of the observed improvements over the Convolutional Neural Operator and standard autoencoder baselines. We will also note that ablation studies varying PDE fidelity are reported in the main text. These additions will allow readers to assess immediately whether the latent PDE evolution or the decoder is the dominant contributor to reconstruction quality. revision: yes

  2. Referee: [—] The manuscript states that 'the fidelity of the embedded PDE relative to the true physics governs the learned latent representation and reconstruction accuracy,' yet provides no independent verification (e.g., comparison of latent trajectories to projected high-fidelity solutions) that the jointly trained encoder produces a representation whose coarse-grid evolution under the prescribed PDE remains close to the true projected dynamics rather than being corrected downstream.

    Authors: We acknowledge that a direct comparison between the evolved latent trajectories and the coarse-grid projections of the high-fidelity solutions would provide stronger evidence that the encoder learns dynamics consistent with the prescribed PDE rather than deferring corrections to the decoder. While the current experiments show that reconstruction accuracy improves monotonically with the fidelity of the embedded PDE, we agree this is indirect. In the revision we will add a dedicated analysis (new figure and accompanying text) that projects the high-fidelity fields onto the latent grid and compares the resulting trajectories with those obtained by evolving the encoder output under the latent PDE. Quantitative trajectory-error metrics and visualizations will be included to address this point explicitly. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation chain is self-contained with external baselines

full rationale

The paper presents DIANO as a novel autoencoding neural operator that encodes high-dimensional fields to a coarse latent grid, evolves a user-prescribed low-fidelity parametric PDE (advection-diffusion or Pressure-Poisson) on that grid via a differentiable solver, and decodes back to the original resolution. The abstract and described framework explicitly compare performance and latent structures against independent external baselines (Convolutional Neural Operator, standard autoencoders) on benchmark flows. No quoted equation, fitted parameter, or self-citation is shown to define the target reconstruction accuracy or latent fidelity by construction; the claim that embedded-PDE fidelity governs reconstruction quality is presented as an experimental observation rather than a definitional identity. The architecture therefore remains non-circular under the stated criteria.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

The central claim rests on the ability of neural operators to perform accurate spatial coarsening and refinement while preserving enough information for a coarse PDE to approximate the original dynamics; several architectural and discretization choices function as free parameters.

free parameters (2)
  • latent grid resolution / coarsening factor
    The degree of spatial reduction is chosen to trade off dimensionality reduction against fidelity and must be tuned for each problem.
  • neural operator architecture hyperparameters
    Network widths, depths, and kernel choices are free parameters that control the encoding and decoding operators.
axioms (2)
  • domain assumption Neural operators can learn accurate mappings between function spaces on different discretizations.
    Invoked when the encoder maps high-resolution inputs to the coarse latent grid and the decoder maps back.
  • domain assumption A discretized version of the governing PDE remains a useful approximation when solved on the coarse latent grid.
    Central premise that allows the differentiable PDE solver to be placed inside the latent space.
invented entities (1)
  • latent-space PDE operator no independent evidence
    purpose: Serves as the sole input-output mapping inside the reduced representation to enforce physics during training and inference.
    The operator is constructed to approximate the original dynamics but is not independently verified outside the training loop.

pith-pipeline@v0.9.0 · 5820 in / 1609 out tokens · 64330 ms · 2026-05-18T11:22:17.636923+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. SPAMoE: Spectrum-Aware Hybrid Operator Framework for Full-Waveform Inversion

    cs.LG 2026-04 unverdicted novelty 7.0

    SPAMoE reduces average MAE by 44.4% on OpenFWI datasets for full-waveform inversion via a spectral-preserving DINO encoder and dynamic frequency-band routing to specialized neural operators.

Reference graph

Works this paper leans on

95 extracted references · 95 canonical work pages · cited by 1 Pith paper · 2 internal anchors

  1. [1]

    J. L. Lumley, The structure of inhomogeneous turbulent flows, Atmospheric turbulence and radio wave propagation (1967) 166–178

    work page 1967

  2. [2]

    Berkooz, P

    G. Berkooz, P. Holmes, J. L. Lumley, The proper orthogonal decomposition in the analysis of turbulent flows, Annual review of fluid mechanics 25 (1) (1993) 539–575

    work page 1993

  3. [3]

    P. J. Schmid, Dynamic mode decomposition of numerical and experimental data, Journal of fluid mechanics 656 (2010) 5–28

    work page 2010

  4. [4]

    Z. J. Zhang, K. Duraisamy, Machine learning methods for data-driven turbulence modeling, in: 22nd AIAA computational fluid dynamics conference, 2015, p. 2460

    work page 2015

  5. [5]

    J. Ling, A. Kurzawski, J. Templeton, Reynolds averaged turbulence modelling using deep neural networks with embedded invariance, Journal of Fluid Mechanics 807 (2016) 155–166

    work page 2016

  6. [6]

    J. N. Kutz, Deep learning in fluid dynamics, Journal of Fluid Mechanics 814 (2017) 1–4

    work page 2017

  7. [7]

    Eivazi, H

    H. Eivazi, H. Veisi, M. H. Naderi, V. Esfahanian, Deep neural networks for nonlinear model order reduction of unsteady flows, Physics of Fluids 32 (10) (2020). p. 29

    work page 2020

  8. [8]

    S. T. Roweis, L. K. Saul, Nonlinear dimensionality reduction by locally linear embedding, science 290 (5500) (2000) 2323–2326

    work page 2000

  9. [9]

    Csala, S

    H. Csala, S. Dawson, A. Arzani, Comparing different nonlinear dimensionality reduction tech- niques for data-driven unsteady fluid flow modeling, Physics of Fluids 34 (11) (2022)

    work page 2022

  10. [10]

    R. Li, S. Song, Manifold learning-based reduced-order model for full speed flow field, Physics of Fluids 36 (8) (2024)

    work page 2024

  11. [11]

    N. B. Erichson, L. Mathelin, Z. Yao, S. L. Brunton, M. W. Mahoney, J. N. Kutz, Shallow neural networks for fluid flow reconstruction with limited sensors, Proceedings of the Royal Society A 476 (2238) (2020) 20200097

    work page 2020

  12. [12]

    Agostini, Exploration and prediction of fluid dynamical systems using auto-encoder tech- nology, Physics of Fluids 32 (6) (2020)

    L. Agostini, Exploration and prediction of fluid dynamical systems using auto-encoder tech- nology, Physics of Fluids 32 (6) (2020)

    work page 2020

  13. [13]

    Fukami, T

    K. Fukami, T. Nakamura, K. Fukagata, Convolutional neural network based hierarchical au- toencoder for nonlinear mode decomposition of fluid field data, Physics of Fluids 32 (9) (2020)

    work page 2020

  14. [14]

    Fukami, K

    K. Fukami, K. Hasegawa, T. Nakamura, M. Morimoto, K. Fukagata, Model order reduction with neural networks: Application to laminar and turbulent flows, SN Computer Science 2 (6) (2021) 467

    work page 2021

  15. [15]

    Nakamura, K

    T. Nakamura, K. Fukami, K. Hasegawa, Y. Nabae, K. Fukagata, Convolutional neural network and long short-term memory based reduced order surrogate for minimal turbulent channel flow, Physics of Fluids 33 (2) (2021)

    work page 2021

  16. [16]

    Sakurada, T

    M. Sakurada, T. Yairi, Anomaly detection using autoencoders with nonlinear dimensionality reduction, in: Proceedings of the MLSDA 2014 2nd workshop on machine learning for sensory data analysis, 2014, pp. 4–11

    work page 2014

  17. [17]

    Cheng, F

    M. Cheng, F. Fang, C. Pain, I. Navon, An advanced hybrid deep adversarial autoencoder for parameterized nonlinear fluid flow modelling, Computer Methods in Applied Mechanics and Engineering 372 (2020) 113375

    work page 2020

  18. [18]

    J. Qu, W. Cai, Y. Zhao, Deep learning method for identifying the minimal representations and nonlinear mode decomposition of fluid flows, Physics of Fluids 33 (10) (2021)

    work page 2021

  19. [19]

    Q. Cao, S. Goswami, G. E. Karniadakis, Laplace neural operator for solving differential equa- tions, Nature Machine Intelligence 6 (6) (2024) 631–640

    work page 2024

  20. [20]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, A. Anand- kumar, Fourier neural operator for parametric partial differential equations, arXiv preprint arXiv:2010.08895 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  21. [21]

    Raonic, R

    B. Raonic, R. Molinaro, T. Rohner, S. Mishra, E. de Bezenac, Convolutional neural operators, in: ICLR 2023 workshop on physics for machine learning, 2023

    work page 2023

  22. [22]

    L. Lu, P. Jin, G. E. Karniadakis, Deeponet: Learning nonlinear operators for identifying dif- ferential equations based on the universal approximation theorem of operators, arXiv preprint arXiv:1910.03193 (2019). p. 30

    work page internal anchor Pith review Pith/arXiv arXiv 1910

  23. [23]

    L. Lu, X. Meng, S. Cai, Z. Mao, S. Goswami, Z. Zhang, G. E. Karniadakis, A comprehensive and fair comparison of two neural operators (with practical extensions) based on fair data, Computer Methods in Applied Mechanics and Engineering 393 (2022) 114778

    work page 2022

  24. [24]

    M. A. Rahman, M. A. Florez, A. Anandkumar, Z. E. Ross, K. Azizzadenesheli, Generative adversarial neural operators, arXiv preprint arXiv:2205.03017 (2022)

    work page arXiv 2022

  25. [25]

    J. H. Lim, N. B. Kovachki, R. Baptista, C. Beckham, K. Azizzadenesheli, J. Kossaifi, V. Voleti, J.Song, K.Kreis, J.Kautz, etal., Score-baseddiffusionmodelsinfunctionspace, arXivpreprint arXiv:2302.07400 (2023)

    work page arXiv 2023

  26. [26]

    J. H. Seidman, G. Kissas, G. J. Pappas, P. Perdikaris, Variational autoencoding neural opera- tors, arXiv preprint arXiv:2302.10351 (2023)

    work page arXiv 2023

  27. [27]

    Kovachki, Z

    N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, A. Anandkumar, Neural operator: Learning maps between function spaces with applications to PDEs, Journal of Machine Learning Research 24 (89) (2023) 1–97

    work page 2023

  28. [28]

    Azizzadenesheli, N

    K. Azizzadenesheli, N. Kovachki, Z. Li, M. Liu-Schiaffini, J. Kossaifi, A. Anandkumar, Neural operators for accelerating scientific simulations and design, Nature Reviews Physics 6 (5) (2024) 320–328

    work page 2024

  29. [29]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, A. Stuart, K. Bhattacharya, A. Anandku- mar, Multipole graph neural operator for parametric partial differential equations, Advances in Neural Information Processing Systems 33 (2020) 6755–6766

    work page 2020

  30. [30]

    Z. Li, K. Meidani, A. B. Farimani, Transformer for partial differential equations’ operator learning, arXiv preprint arXiv:2205.13671 (2022)

    work page arXiv 2022

  31. [31]

    Serrano, L

    L. Serrano, L. Le Boudec, A. Kassaï Koupaï, T. X. Wang, Y. Yin, J.-N. Vittaut, P. Gallinari, Operator learning with neural fields: Tackling PDEs on general geometries, Advances in Neural Information Processing Systems 36 (2023) 70581–70611

    work page 2023

  32. [32]

    Z. Hao, Z. Wang, H. Su, C. Ying, Y. Dong, S. Liu, Z. Cheng, J. Song, J. Zhu, GNOT: A general neural operator transformer for operator learning, in: International Conference on Machine Learning, PMLR, 2023, pp. 12556–12569

    work page 2023

  33. [33]

    Alkin, A

    B. Alkin, A. Fürst, S. Schmid, L. Gruber, M. Holzleitner, J. Brandstetter, Universal physics transformers: A framework for efficiently scaling neural operators, Advances in Neural Infor- mation Processing Systems 37 (2024) 25152–25194

    work page 2024

  34. [34]

    Z. Li, N. Kovachki, C. Choy, B. Li, J. Kossaifi, S. Otta, M. A. Nabian, M. Stadler, C. Hundt, K. Azizzadenesheli, et al., Geometry-informed neural operator for large-scale 3D PDEs, Ad- vances in Neural Information Processing Systems 36 (2023) 35836–35854

    work page 2023

  35. [35]

    X. Han, J. Zhang, D. Samaras, F. Hou, H. Qin, GeoMaNO: Geometric Mamba Neural Operator for Partial Differential Equations, arXiv preprint arXiv:2505.12020 (2025)

    work page arXiv 2025

  36. [36]

    S. R. Bukka, R. Gupta, A. R. Magee, R. K. Jaiman, Assessment of unsteady flow predictions using hybrid deep learning based reduced-order models, Physics of Fluids 33 (1) (2021). p. 31

    work page 2021

  37. [37]

    Gupta, R

    R. Gupta, R. Jaiman, Three-dimensional deep learning-based reduced order model for unsteady flow dynamics with variable Reynolds number, Physics of Fluids 34 (3) (2022)

    work page 2022

  38. [38]

    R. T. Chen, Y. Rubanova, J. Bettencourt, D. K. Duvenaud, Neural ordinary differential equa- tions, Advances in neural information processing systems 31 (2018)

    work page 2018

  39. [39]

    Zhang, Nonlinear mode decomposition via physics-assimilated convolutional autoencoder for unsteady flows over an airfoil, Physics of Fluids 35 (9) (2023)

    B. Zhang, Nonlinear mode decomposition via physics-assimilated convolutional autoencoder for unsteady flows over an airfoil, Physics of Fluids 35 (9) (2023)

    work page 2023

  40. [40]

    W. Peng, Z. Yuan, J. Wang, Attention-enhanced neural network models for turbulence simu- lation, Physics of Fluids 34 (2) (2022)

    work page 2022

  41. [41]

    Tripura, S

    T. Tripura, S. Chakraborty, Wavelet neural operator: a neural operator for parametric partial differential equations, arXiv preprint arXiv:2205.02191 (2022)

    work page arXiv 2022

  42. [42]

    M. A. Rahman, Z. E. Ross, K. Azizzadenesheli, U-NO: U-shaped neural operators, arXiv preprint arXiv:2204.11127 (2022)

    work page arXiv 2022

  43. [43]

    G. Chen, X. Liu, Q. Meng, L. Chen, C. Liu, Y. Li, Learning neural operators on riemannian manifolds, National Science Open 3 (6) (2024) 20240001

    work page 2024

  44. [44]

    X. Fu, G. Chen, Y. Li, X. Liu, L. Chen, Q. Meng, C. Liu, X. Hao, Spatio-temporal neural operator on complex geometries, Computer Physics Communications (2025) 109754

    work page 2025

  45. [45]

    W. Peng, Z. Yuan, Z. Li, J. Wang, Linear attention coupled fourier neural operator for simu- lation of three-dimensional turbulence, Physics of Fluids 35 (1) (2023)

    work page 2023

  46. [46]

    Ye, C.-S

    Z. Ye, C.-S. Zhang, W. Wang, Recurrent Neural Operators: Stable Long-Term PDE Prediction, arXiv preprint arXiv:2505.20721 (2025)

    work page arXiv 2025

  47. [47]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics-informed machine learning, Nature Reviews Physics 3 (6) (2021) 422–440

    work page 2021

  48. [48]

    S. A. Faroughi, N. Pawar, C. Fernandes, M. Raissi, S. Das, N. K. Kalantari, S. K. Mahjour, Physics-guided, physics-informed, and physics-encoded neural networks in scientific computing, arXiv preprint arXiv:2211.07377 (2022)

    work page arXiv 2022

  49. [49]

    Raissi, P

    M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, Journal of Computational physics 378 (2019) 686–707

    work page 2019

  50. [50]

    Mattey, S

    R. Mattey, S. Ghosh, A physics informed neural network for time-dependent nonlinear and higher order partial differential equations, arXiv preprint arXiv:2106.07606 (2021)

    work page arXiv 2021

  51. [51]

    Arzani, K

    A. Arzani, K. W. Cassel, R. M. D’Souza, Theory-guided physics-informed neural networks for boundary layer problems with singular perturbation, Journal of Computational Physics 473 (2023) 111768

    work page 2023

  52. [52]

    A. T. Mohan, N. Lubbers, M. Chertkov, D. Livescu, Embedding hard physical constraints in neural network coarse-graining of three-dimensional turbulence, Physical Review Fluids 8 (1) (2023) 014604. p. 32

    work page 2023

  53. [53]

    Chalapathi, Y

    N. Chalapathi, Y. Du, A. Krishnapriyan, Scaling physics-informed hard constraints with mixture-of-experts, arXiv preprint arXiv:2402.13412 (2024)

    work page arXiv 2024

  54. [54]

    Karnakov, S

    P. Karnakov, S. Litvinov, P. Koumoutsakos, Solving inverse problems in physics by optimizing a discrete loss: Fast and accurate learning without neural networks, PNAS nexus 3 (1) (2024) pgae005

    work page 2024

  55. [55]

    Wiewel, M

    S. Wiewel, M. Becher, N. Thuerey, Latent space physics: Towards learning the temporal evolution of fluid flow, in: Computer graphics forum, Vol. 38, Wiley Online Library, 2019, pp. 71–82

    work page 2019

  56. [56]

    T. Wu, T. Maruyama, J. Leskovec, Learning to accelerate partial differential equations via latent global evolution, Advances in Neural Information Processing Systems 35 (2022) 2240– 2253

    work page 2022

  57. [57]

    Message passing neural pde solvers.arXiv preprint arXiv:2202.03376, 2022

    J. Brandstetter, D. Worrall, M. Welling, Message passing neural PDE solvers, arXiv preprint arXiv:2202.03376 (2022)

    work page arXiv 2022

  58. [58]

    P.Lippe, B.Veeling, P.Perdikaris, R.Turner, J.Brandstetter, PDE-refiner: Achievingaccurate long rollouts with neural PDE solvers, Advances in Neural Information Processing Systems 36 (2023) 67398–67433

    work page 2023

  59. [59]

    X.-Y. Liu, M. Zhu, L. Lu, H. Sun, J.-X. Wang, Multi-resolution partial differential equations preserved learning framework for spatiotemporal dynamics, Communications Physics 7 (1) (2024) 31

    work page 2024

  60. [60]

    Z. Li, S. Patil, F. Ogoke, D. Shu, W. Zhen, M. Schneier, J. R. Buchanan Jr, A. B. Farimani, La- tent neural PDE solver: A reduced-order modeling framework for partial differential equations, Journal of Computational Physics 524 (2025) 113705

    work page 2025

  61. [61]

    Boral, Z

    A. Boral, Z. Y. Wan, L. Zepeda-Núñez, J. Lottes, Q. Wang, Y.-f. Chen, J. Anderson, F. Sha, Neural ideal large eddy simulation: Modeling turbulence with neural stochastic differential equations, Advances in neural information processing systems 36 (2023) 69270–69283

    work page 2023

  62. [62]

    Sirignano, J

    J. Sirignano, J. F. MacArt, Dynamic deep learning les closures: Online optimization with embedded dns, arXiv preprint arXiv:2303.02338 (2023)

    work page arXiv 2023

  63. [63]

    Kochkov, J

    D. Kochkov, J. A. Smith, A. Alieva, Q. Wang, M. P. Brenner, S. Hoyer, Machine learning– accelerated computational fluid dynamics, Proceedings of the National Academy of Sciences 118 (21) (2021) e2101784118

    work page 2021

  64. [64]

    Fan, J.-X

    X. Fan, J.-X. Wang, Differentiable hybrid neural modeling for fluid-structure interaction, Jour- nal of Computational Physics 496 (2024) 112584

    work page 2024

  65. [65]

    Tompson, K

    J. Tompson, K. Schlachter, P. Sprechmann, K. Perlin, Accelerating eulerian fluid simulation with convolutional networks, in: International conference on machine learning, PMLR, 2017, pp. 3424–3433

    work page 2017

  66. [66]

    Duraisamy, G

    K. Duraisamy, G. Iaccarino, H. Xiao, Turbulence modeling in the age of data, Annual review of fluid mechanics 51 (1) (2019) 357–377. p. 33

    work page 2019

  67. [67]

    Vinuesa, S

    R. Vinuesa, S. L. Brunton, Enhancing computational fluid dynamics with machine learning, Nature Computational Science 2 (6) (2022) 358–366

    work page 2022

  68. [68]

    Margenberg, R

    N. Margenberg, R. Jendersie, C. Lessig, T. Richter, DNN-MG: A hybrid neural network/finite element method with applications to 3D simulations of the Navier–Stokes equations, Computer Methods in Applied Mechanics and Engineering 420 (2024) 116692

    work page 2024

  69. [69]

    F. D. A. Belbute-Peres, T. Economon, Z. Kolter, Combining differentiable pde solvers and graph neural networks for fluid flow prediction, in: international conference on machine learn- ing, PMLR, 2020, pp. 2402–2411

    work page 2020

  70. [70]

    List, L.-W

    B. List, L.-W. Chen, N. Thuerey, Learned turbulence modelling with differentiable fluid solvers: physics-based loss functions and optimisation horizons, Journal of Fluid Mechanics 949 (2022) A25

    work page 2022

  71. [71]

    List, L.-W

    B. List, L.-W. Chen, K. Bali, N. Thuerey, Differentiability in unrolled training of neural physics simulators on transient dynamics, Computer Methods in Applied Mechanics and Engineering 433 (2025) 117441

    work page 2025

  72. [72]

    Akhare, P

    D. Akhare, P. Du, T. Luo, J.-X. Wang, Implicit neural differential model for spatiotemporal dynamics, arXiv preprint arXiv:2504.02260 (2025)

    work page arXiv 2025

  73. [73]

    C. Wang, J. Berner, Z. Li, D. Zhou, J. Wang, J. Bae, A. Anandkumar, Beyond clo- sure models: Learning chaotic-systems via physics-informed neural operators, arXiv preprint arXiv:2408.05177 (2024)

    work page arXiv 2024

  74. [74]

    W. D. Fries, X. He, Y. Choi, LaSDI: Parametric latent space dynamics identification, Computer Methods in Applied Mechanics and Engineering 399 (2022) 115436

    work page 2022

  75. [75]

    Champion, B

    K. Champion, B. Lusch, J. N. Kutz, S. L. Brunton, Data-driven discovery of coordinates and governing equations, Proceedings of the National Academy of Sciences 116 (45) (2019) 22445– 22451

    work page 2019

  76. [76]

    P. A. Reinbold, R. O. Grigoriev, Data-driven discovery of partial differential equation models with latent variables, Physical Review E 100 (2) (2019) 022219

    work page 2019

  77. [77]

    Maulik, A

    R. Maulik, A. Mohan, B. Lusch, S. Madireddy, P. Balaprakash, D. Livescu, Time-series learning of latent-space dynamics for reduced-order model closure, Physica D: Nonlinear Phenomena 405 (2020) 132368

    work page 2020

  78. [78]

    D. Z. Huang, K. Xu, C. Farhat, E. Darve, Learning constitutive relations from indirect obser- vations using deep neural networks, Journal of Computational Physics 416 (2020) 109491

    work page 2020

  79. [79]

    Négiar, M

    G. Négiar, M. W. Mahoney, A. S. Krishnapriyan, Learning differentiable solvers for systems with hard constraints, arXiv preprint arXiv:2207.08675 (2022)

    work page arXiv 2022

  80. [80]

    J. S. R. Park, S. W. Cheung, Y. Choi, Y. Shin, tLaSDI: Thermodynamics-informed latent space dynamics identification, Computer Methods in Applied Mechanics and Engineering 429 (2024) 117144. p. 34

    work page 2024

Showing first 80 references.