pith. sign in

arxiv: 2606.09044 · v1 · pith:ISZSEOWQnew · submitted 2026-06-08 · ⚛️ physics.flu-dyn

A Hybrid Generative Reduced-Order Model for the Minimal Flow Unit

Niccol\`o Tonioni , Lionel Agostini , Marcial Sanchis-Agudo , Mohammad Umair , Franck Kerherv\'e , Laurent Cordier , Ricardo Vinuesa This is my paper

Pith reviewed 2026-06-27 15:20 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords reduced-order modelingwall-bounded turbulencevariational autoencodertransformerminimal flow unitturbulent kinetic energynear-wall regeneration cyclegenerative model
0
0 comments X

The pith

A hybrid β-VAE-GAN and sensor-conditioned Transformer compresses wall-bounded turbulence into four latent dimensions that recover 87 percent of turbulent kinetic energy and forecast intermittent dynamics from sparse sensors.

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

The paper develops a data-driven reduced-order model that first uses a β-VAE-GAN to map high-dimensional flow fields from the minimal flow unit into a compact latent space. A Transformer module conditioned on sensor data then predicts the time evolution of those latent variables using an efficient static attention operator. On the minimal flow unit at Reynolds number 200 and wall distance y+ = 14, the latent representation captures 87 percent of the turbulent kinetic energy and autonomously encodes the low-frequency timescale of the near-wall regeneration cycle. The full pipeline sustains accurate long-horizon forecasts while reproducing the cycle's active and quiescent phases.

Core claim

The hybrid framework compresses the Minimal Flow Unit flow at Reτ = 200 and y+ = 14 into a four-dimensional latent space recovering 87 percent of the turbulent kinetic energy, exceeding a standard β-VAE baseline by more than 10 percent. The latent coordinates autonomously encode the characteristic timescales of the flow, with specific dimensions capturing the low-frequency signature of the near-wall regeneration cycle at T+ ≈ 1724. The sensor-conditioned Transformer using Easy Attention and AdaLN-Zero maintains accurate forecasts over 17,288 t+ from an initialisation window of only 128 t+, while end-to-end inference reconstructs 82 percent of the turbulent kinetic energy and reproduces the a

What carries the argument

The β-VAE-GAN compression stage paired with a sensor-conditioned Transformer that replaces standard self-attention with Easy Attention and applies AdaLN-Zero modulation for sensor conditioning.

If this is right

  • The four latent dimensions autonomously encode the low-frequency signature of the near-wall regeneration cycle at T+ ≈ 1724.
  • Forecasts remain accurate for 17,288 time units when initialised from only 128 time units of sensor data.
  • End-to-end inference from the full pipeline reconstructs 82 percent of the turbulent kinetic energy.
  • The model reproduces the intermittent active and quiescent phases of the regeneration cycle despite attenuating rare extreme events.

Where Pith is reading between the lines

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

  • The approach could reduce the computational cost of simulating wall-bounded turbulence in engineering design by replacing full direct numerical simulations with latent-space evolution.
  • Testing the same architecture on minimal flow units at higher Reynolds numbers would reveal whether the four-dimensional bottleneck remains sufficient.
  • The autonomous encoding of specific timescales suggests the latent space might be used to isolate and study other hidden periodicities in turbulent data.

Load-bearing premise

The encoder's focus on statistically recurrent flow states does not prevent accurate reproduction of the alternating active and quiescent phases of the regeneration cycle.

What would settle it

A direct numerical simulation of the minimal flow unit at the same Reynolds number and wall location, followed by comparison of the model's predicted regeneration-cycle period against the simulation's measured period; a large mismatch would falsify the claim of suitability as a surrogate.

Figures

Figures reproduced from arXiv: 2606.09044 by Franck Kerherv\'e, Laurent Cordier, Lionel Agostini, Marcial Sanchis-Agudo, Mohammad Umair, Niccol\`o Tonioni, Ricardo Vinuesa.

Figure 1
Figure 1. Figure 1: Schematic representation of the proposed model framework. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Instantaneous visualization of the MFU computational domain. Iso-surfaces of the Q-criterion [ [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of the datasets first- and second-order statistics against reference DNS data. Left panel: [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Schematic of the β-VAE-GAN architecture. The encoder (E), generator (G), and discriminator (D) utilize convolutional layers. Each ConvCf layer applies k × k filters with Cf output channels. Downward arrows (↓ (2, 1)) denote strided downsampling in the streamwise and spanwise directions, respectively, while upward arrows (↑ (2, 1)) denote Lanczos upsampling. Layers labeled Conv Cµ and Conv Cσ project the en… view at source ↗
Figure 5
Figure 5. Figure 5: Schematic of the sensor-conditioned Transformer architecture. The framework processes sequences of latent [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of compression performance on the test set across different latent dimensions [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Two-point spatial autocorrelation of the streamwise velocity fluctuations, [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Temporal evolution of the four latent variables over an extended autoregressive horizon. Left panels: time [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Spectral and statistical analysis of the four latent variables. Left panel: power spectral density (PSD) of the [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Temporal evolution and associated power spectral densities (PSDs) of selected physical flow statistics. Left [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Two-point spatial autocorrelation of the streamwise ( [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Singular value decay of the POD modes for the DNS test data (gray), the compression baseline (orange), and [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Quadrant analysis of the Reynolds shear stress. Top row: joint probability density function [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Temporal evolution of the spatial maximum of the instantaneous Reynolds shear stress, [PITH_FULL_IMAGE:figures/full_fig_p023_14.png] view at source ↗
read the original abstract

A data-driven reduced-order modelling framework is proposed for wall-bounded turbulent flows to forecast the intermittent near-wall dynamics over extended time horizons from sparse sensor measurements. The approach combines a $\beta$-VAE-GAN, which compresses high-dimensional flow fields into a low-dimensional latent space, with a sensor-conditioned Transformer that forecasts the evolution of the latent variables. The temporal module employs Easy Attention, a static time-mixing operator that replaces the learnable query-key mechanism of standard self-attention at reduced computational cost, combined with an adapted AdaLN-Zero modulation mechanism for sensor-based conditioning. Evaluated on the Minimal Flow Unit ($Re_\tau = 200$) at $y^+ = 14$, the compression stage recovers $87\%$ of the turbulent kinetic energy within a four-dimensional latent space, exceeding the standard $\beta$-VAE baseline by more than $10\%$. The latent dimensions autonomously encode the characteristic timescales of the flow, with specific coordinates capturing the low-frequency signature of the near-wall regeneration cycle ($T^+ \approx 1724$), establishing the physical interpretability of the learnt representation. The sensor-conditioned Transformer maintains accurate forecasts over $17{,}288\,t^+$ from an initialisation window of only $128\,t^+$, whilst end-to-end inference reconstructs $82\%$ of the turbulent kinetic energy. The principal limitation is the attenuation of rare, extreme-amplitude events, a consequence of the encoder prioritising the most statistically recurrent flow states within the low-dimensional bottleneck. Nevertheless, the framework accurately reproduces the alternating active and quiescent phases of the regeneration cycle, demonstrating its suitability as a surrogate model for the intermittent dynamics of wall-bounded turbulence.

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 a hybrid generative reduced-order model for wall-bounded turbulence in the Minimal Flow Unit (Re_τ=200, y+=14), combining a β-VAE-GAN to compress flow fields into a 4D latent space (recovering 87% TKE, >10% above standard β-VAE) with a sensor-conditioned Transformer using Easy Attention and AdaLN-Zero for forecasting latent dynamics from sparse sensors. It claims accurate forecasts over 17,288 t+ from a 128 t+ window, end-to-end reconstruction of 82% TKE, autonomous encoding of flow timescales including the near-wall regeneration cycle (T+≈1724), and suitability as a surrogate despite attenuating rare extreme events while reproducing active/quiescent phases.

Significance. If the performance metrics and surrogate suitability hold under scrutiny, the work demonstrates a computationally efficient data-driven framework for long-horizon forecasting of intermittent near-wall turbulence from limited sensors, with notable physical interpretability arising from the latent space autonomously capturing characteristic timescales.

major comments (2)
  1. [Abstract] Abstract: The central claim of suitability as a surrogate model for intermittent near-wall dynamics rests on the assertion that reproduction of alternating active and quiescent phases (with T+≈1724 signature) is sufficient despite acknowledged attenuation of rare extreme-amplitude events. However, the regeneration cycle is defined by the statistics of those extremes; no quantitative evidence (e.g., duty cycle, transition probabilities, or conditional spectra) is provided to confirm that phase alternation alone preserves the cycle's defining properties without bias from the encoder's prioritization of recurrent states.
  2. [Abstract] Abstract: The concrete performance claims (87% TKE recovery in 4D latent space, 17,288 t+ forecast horizon, 82% end-to-end TKE) are presented without reference to training details, validation splits, error bars, ablation studies, or sensitivity to the free parameters (latent dimension, β, Transformer hyperparameters), rendering the metrics unverifiable and the robustness of the compression and forecasting stages unassessed.
minor comments (2)
  1. [Abstract] Abstract: Inconsistent comma formatting in large numbers (17,288 vs 17{,}288) should be standardized for clarity.
  2. The description of 'Easy Attention' as a static time-mixing operator replacing learnable query-key mechanisms would benefit from a brief equation or pseudocode reference to distinguish it from standard self-attention.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation and strengthen the claims regarding surrogate suitability and metric robustness. We address each point below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of suitability as a surrogate model for intermittent near-wall dynamics rests on the assertion that reproduction of alternating active and quiescent phases (with T+≈1724 signature) is sufficient despite acknowledged attenuation of rare extreme-amplitude events. However, the regeneration cycle is defined by the statistics of those extremes; no quantitative evidence (e.g., duty cycle, transition probabilities, or conditional spectra) is provided to confirm that phase alternation alone preserves the cycle's defining properties without bias from the encoder's prioritization of recurrent states.

    Authors: We agree that the current evidence for surrogate suitability relies primarily on the observed reproduction of active/quiescent alternation and the autonomous capture of the T+≈1724 timescale in the latent space. While the manuscript notes the attenuation of extremes as a limitation, it does not provide the requested quantitative metrics such as duty cycle, transition probabilities, or conditional spectra. In the revised manuscript we will add these analyses (computed from the latent trajectories and reconstructed fields) to demonstrate that the phase statistics remain unbiased relative to the full DNS. revision: yes

  2. Referee: [Abstract] Abstract: The concrete performance claims (87% TKE recovery in 4D latent space, 17,288 t+ forecast horizon, 82% end-to-end TKE) are presented without reference to training details, validation splits, error bars, ablation studies, or sensitivity to the free parameters (latent dimension, β, Transformer hyperparameters), rendering the metrics unverifiable and the robustness of the compression and forecasting stages unassessed.

    Authors: We acknowledge that the abstract states the headline metrics without accompanying methodological qualifiers. The full manuscript contains training protocols, data splits, and some hyperparameter choices in the Methods section, but lacks error bars, systematic ablations, and sensitivity sweeps. The revised manuscript will expand the abstract (or add a short “Performance Summary” paragraph) to reference these details, include error bars from multiple random seeds, and report ablation results on latent dimension, β, and key Transformer hyperparameters. revision: yes

Circularity Check

0 steps flagged

No significant circularity; purely empirical evaluation on held-out data

full rationale

The paper describes a data-driven pipeline (β-VAE-GAN compression followed by sensor-conditioned Transformer forecasting) whose performance metrics (87% TKE recovery in 4D latent space, 82% end-to-end reconstruction, forecasts to 17,288 t+) are obtained by training networks on simulation snapshots and evaluating on held-out data. No load-bearing step reduces by the paper's own equations or self-citations to a fitted parameter renamed as a prediction; the latent-space interpretability claims are post-hoc observations rather than derivations. The framework is therefore self-contained against external benchmarks with no circular reduction.

Axiom & Free-Parameter Ledger

3 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard machine-learning assumptions plus several fitted architectural choices whose values are not reported; no new physical entities are postulated.

free parameters (3)
  • latent dimension
    Set to 4; chosen to balance compression and reconstruction quality on the training data.
  • β in β-VAE
    Regularization weight in the variational loss; value not stated but required for the reported 87 % recovery.
  • Transformer hyperparameters
    Number of layers, heads, and Easy Attention parameters fitted during training.
axioms (2)
  • domain assumption The training and test snapshots are statistically representative of the Minimal Flow Unit dynamics at y+ = 14.
    Invoked implicitly when claiming long-horizon forecast accuracy and physical interpretability of latent coordinates.
  • ad hoc to paper The low-dimensional bottleneck preserves the essential intermittent structure even while attenuating extremes.
    Required for the suitability claim despite the stated limitation on rare events.

pith-pipeline@v0.9.1-grok · 5866 in / 1754 out tokens · 33367 ms · 2026-06-27T15:20:11.027028+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

68 extracted references · 2 canonical work pages

  1. [1]

    Near-wall turbulence.Physics of Fluids, 25(10), 2013

    Javier Jiménez. Near-wall turbulence.Physics of Fluids, 25(10), 2013

    2013

  2. [2]

    On the influence of outer large-scale structures on near-wall turbulence in channel flow.Physics of Fluids, 26(7), 2014

    L Agostini and MA Leschziner. On the influence of outer large-scale structures on near-wall turbulence in channel flow.Physics of Fluids, 26(7), 2014

    2014

  3. [3]

    The local structure of turbulence in incompressible viscous fluid for very large reynolds.Numbers

    Andrey Nikolaevich Kolmogorov. The local structure of turbulence in incompressible viscous fluid for very large reynolds.Numbers. In Dokl. Akad. Nauk SSSR, 30:301, 1941

    1941

  4. [4]

    Springer Berlin Heidelberg, Berlin, Heidelberg, 2003

    Luca Biferale, Guido Boffetta, and Bernard Castaing.Fully Developed Turbulence, pages 149–172. Springer Berlin Heidelberg, Berlin, Heidelberg, 2003

    2003

  5. [5]

    Coherent motions in the turbulent boundary layer.Annual review of fluid mechanics, 23(1): 601–639, 1991

    Stephenk Robinson. Coherent motions in the turbulent boundary layer.Annual review of fluid mechanics, 23(1): 601–639, 1991

    1991

  6. [6]

    Large-scale amplitude modulation of the small-scale structures in turbulent boundary layers.Journal of Fluid Mechanics, 628:311–337, 2009

    Romain Mathis, Nicholas Hutchins, and Ivan Marusic. Large-scale amplitude modulation of the small-scale structures in turbulent boundary layers.Journal of Fluid Mechanics, 628:311–337, 2009

    2009

  7. [7]

    High reynolds number effects in wall turbulence.Interna- tional Journal of Heat and Fluid Flow, 31(3):418–428, 2010

    Ivan Marusic, Romain Mathis, and Nicholas Hutchins. High reynolds number effects in wall turbulence.Interna- tional Journal of Heat and Fluid Flow, 31(3):418–428, 2010

    2010

  8. [8]

    Predictive model for wall-bounded turbulent flow.Science, 329(5988):193–196, 2010

    Ivan Marusic, Romain Mathis, and Nicholas Hutchins. Predictive model for wall-bounded turbulent flow.Science, 329(5988):193–196, 2010

    2010

  9. [9]

    On the nature of turbulence.Les rencontres physiciens-mathématiciens de Strasbourg-RCP25, 12:1–44, 1971

    David Ruelle and Floris Takens. On the nature of turbulence.Les rencontres physiciens-mathématiciens de Strasbourg-RCP25, 12:1–44, 1971

    1971

  10. [10]

    Self-sustaining mechanisms of wall turbulence-a review

    Ronald Panton. Self-sustaining mechanisms of wall turbulence-a review. In37th Aerospace Sciences Meeting and Exhibit, page 552, 1999

    1999

  11. [11]

    Regeneration mechanisms of near-wall turbulence structures

    James M Hamilton, John Kim, and Fabian Waleffe. Regeneration mechanisms of near-wall turbulence structures. Journal of Fluid Mechanics, 287:317–348, 1995

    1995

  12. [12]

    On a self-sustaining process in shear flows.Physics of Fluids, 9(4):883–900, 1997

    Fabian Waleffe. On a self-sustaining process in shear flows.Physics of Fluids, 9(4):883–900, 1997. 24 A Hybrid Generative Reduced-Order ModelA PREPRINT

    1997

  13. [13]

    How streamwise rolls and streaks self-sustain in a shear flow

    Fabian Waleffe and John Kim. How streamwise rolls and streaks self-sustain in a shear flow. ii. In29th AIAA, Fluid Dynamics Conference, page 2997, 1998

    1998

  14. [14]

    A low-dimensional model for turbulent shear flows.New Journal of Physics, 6(1):56, 2004

    Jeff Moehlis, Holger Faisst, and Bruno Eckhardt. A low-dimensional model for turbulent shear flows.New Journal of Physics, 6(1):56, 2004

    2004

  15. [15]

    Genesis and dynamics of coherent structures in near-wall turbulence: a new look.Self-sustaining Mechanisms of wall Turbulence, 1997

    Wade Schoppa. Genesis and dynamics of coherent structures in near-wall turbulence: a new look.Self-sustaining Mechanisms of wall Turbulence, 1997

    1997

  16. [16]

    J. L. Lumley. The structure of inhomogeneous turbulent flows. In A. M. Yaglom and V . I. Tartarsky, editors, Atmospheric Turbulence and Radio Wave Propagation, pages 166–177. Nauka, 1967

    1967

  17. [17]

    Probability theory: foundations, random sequences.(No Title), 1955

    Michel Loève. Probability theory: foundations, random sequences.(No Title), 1955

    1955

  18. [18]

    Laurent Cordier and Michel Bergmann. Proper orthogonal decomposition: an overview.Lecture series 2002-04, 2003-03 and 2008-01 on post-processing of experimental and numerical data, Von Karman Institute for Fluid Dynamics, 2008., pages 46–pages, 2008

    2002

  19. [19]

    Lumley, Gahl Berkooz, and Clarence W

    Philip Holmes, John L. Lumley, Gahl Berkooz, and Clarence W. Rowley.Turbulence, Coherent Structures, Dynamical Systems and Symmetry. Cambridge Monographs on Mechanics. Cambridge University Press, 2 edition, 2012

    2012

  20. [20]

    Turbulence and the dynamics of coherent structures

    Lawrence Sirovich. Turbulence and the dynamics of coherent structures. i. coherent structures.Quarterly of applied mathematics, 45(3):561–571, 1987

    1987

  21. [21]

    L Cordier and Michel Bergmann. Two typical applications of pod: coherent structures eduction and reduced order modelling.Lecture series 2002-04, 2003-03 and 2008-01 on post-processing of experimental and numerical data, Von Karman Institute for Fluid Dynamics, 2008., pages 60–pages, 2008

    2002

  22. [22]

    The dynamics of coherent structures in the wall region of a turbulent boundary layer.Journal of fluid Mechanics, 192:115–173, 1988

    Nadine Aubry, Philip Holmes, John L Lumley, and Emily Stone. The dynamics of coherent structures in the wall region of a turbulent boundary layer.Journal of fluid Mechanics, 192:115–173, 1988

    1988

  23. [23]

    A low-dimensional approach for the minimal flow unit.Journal of Fluid Mechanics, 362:121–155, 1998

    Bérengère Podvin and John Lumley. A low-dimensional approach for the minimal flow unit.Journal of Fluid Mechanics, 362:121–155, 1998

    1998

  24. [24]

    Coherence and chaos in a model of turbulent boundary layer.Physics of Fluids A: Fluid Dynamics, 4(12):2855–2874, 1992

    Xiang Zhou and L Sirovich. Coherence and chaos in a model of turbulent boundary layer.Physics of Fluids A: Fluid Dynamics, 4(12):2855–2874, 1992

    1992

  25. [25]

    Reduced basis methods: Success, limitations and future challenges.arXiv preprint arXiv:1511.02021, 2015

    Mario Ohlberger and Stephan Rave. Reduced basis methods: Success, limitations and future challenges.arXiv preprint arXiv:1511.02021, 2015

    Pith/arXiv arXiv 2015

  26. [26]

    Decay of the kolmogorov n-width for wave problems.Applied Mathematics Letters, 96:216–222, 2019

    Constantin Greif and Karsten Urban. Decay of the kolmogorov n-width for wave problems.Applied Mathematics Letters, 96:216–222, 2019

    2019

  27. [27]

    Breaking the kolmogorov barrier in model reduction of fluid flows.Fluids, 5(1): 26, 2020

    Shady E Ahmed and Omer San. Breaking the kolmogorov barrier in model reduction of fluid flows.Fluids, 5(1): 26, 2020

    2020

  28. [28]

    Breaking the kolmogorov barrier with nonlinear model reduction.Notices of the American Mathematical Society, 69(5):725–733, 2022

    Benjamin Peherstorfer. Breaking the kolmogorov barrier with nonlinear model reduction.Notices of the American Mathematical Society, 69(5):725–733, 2022

    2022

  29. [29]

    MIT press Cambridge, 2016

    Ian Goodfellow, Yoshua Bengio, Aaron Courville, and Yoshua Bengio.Deep learning, volume 1. MIT press Cambridge, 2016

    2016

  30. [30]

    Coding theorems for a discrete source with a fidelity criterion.IRE Nat

    Claude E Shannon et al. Coding theorems for a discrete source with a fidelity criterion.IRE Nat. Conv. Rec, 4 (142-163):1, 1959

    1959

  31. [31]

    On interpretability and proper latent decomposition of autoencoders.arXiv preprint arXiv:2211.08345, 2022

    Luca Magri and Anh Khoa Doan. On interpretability and proper latent decomposition of autoencoders.arXiv preprint arXiv:2211.08345, 2022

    arXiv 2022

  32. [32]

    Machine learning for fluid mechanics.Annual review of fluid mechanics, 52(1):477–508, 2020

    Steven L Brunton, Bernd R Noack, and Petros Koumoutsakos. Machine learning for fluid mechanics.Annual review of fluid mechanics, 52(1):477–508, 2020

    2020

  33. [33]

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

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

    2020

  34. [34]

    Enhancing computational fluid dynamics with machine learning.Nature Computational Science, 2(6):358–366, 2022

    Ricardo Vinuesa and Steven L Brunton. Enhancing computational fluid dynamics with machine learning.Nature Computational Science, 2(6):358–366, 2022

    2022

  35. [35]

    Perspectives on predicting and controlling turbulent flows through deep learning.Physics of Fluids, 36(3), 2024

    Ricardo Vinuesa. Perspectives on predicting and controlling turbulent flows through deep learning.Physics of Fluids, 36(3), 2024

    2024

  36. [36]

    Lionel Agostini and Michael Leschziner. Auto-encoder-assisted analysis of amplitude and wavelength modulation of near-wall turbulence by outer large-scale structures in channel flow at friction reynolds number of 5200.Physics of Fluids, 34(11), 2022. 25 A Hybrid Generative Reduced-Order ModelA PREPRINT

    2022

  37. [37]

    Predicting the wall-shear stress and wall pressure through convolutional neural networks.International Journal of Heat and Fluid Flow, 103:109200, 2023

    Arivazhagan G Balasubramanian, Luca Guastoni, Philipp Schlatter, Hossein Azizpour, and Ricardo Vinuesa. Predicting the wall-shear stress and wall pressure through convolutional neural networks.International Journal of Heat and Fluid Flow, 103:109200, 2023

    2023

  38. [38]

    Dynamics of a data-driven low-dimensional model of turbulent minimal couette flow.Journal of Fluid Mechanics, 973:A42, 2023

    Alec J Linot and Michael D Graham. Dynamics of a data-driven low-dimensional model of turbulent minimal couette flow.Journal of Fluid Mechanics, 973:A42, 2023

    2023

  39. [39]

    Alberto Solera-Rico, Carlos Sanmiguel Vila, Miguel Gómez-López, Yuning Wang, Abdulrahman Almashjary, Scott TM Dawson, and Ricardo Vinuesa.β-variational autoencoders and transformers for reduced-order modelling of fluid flows.Nature Communications, 15(1):1361, 2024

    2024

  40. [40]

    beta-V AE: Learning basic visual concepts with a constrained variational framework

    Irina Higgins, Loic Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick, Shakir Mohamed, and Alexander Lerchner. beta-V AE: Learning basic visual concepts with a constrained variational framework. In International Conference on Learning Representations, 2017

    2017

  41. [41]

    Generative adversarial nets.Advances in neural information processing systems, 27, 2014

    Ian J Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets.Advances in neural information processing systems, 27, 2014

    2014

  42. [42]

    Vivaldy: A hybrid generative reduced-order model for turbulent flows, applied to vortex-induced vibrations.Physical Review Fluids,

    Niccolò Tonioni, Lionel Agostini, Franck Kerhervé, Laurent Cordier, and Ricardo Vinuesa. Vivaldy: A hybrid generative reduced-order model for turbulent flows, applied to vortex-induced vibrations.Physical Review Fluids,

  43. [43]

    Accepted for publication

  44. [44]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 4195–4205, October 2023

    2023

  45. [45]

    Easy attention: A simple attention mechanism for temporal predictions with transformers.APL Computational Physics, 1(1):016104, 2025

    Marcial Sanchis-Agudo, Yuning Wang, Roger Arnau, Luca Guastoni, Jasmin Lim, Karthik Duraisamy, and Ricardo Vinuesa. Easy attention: A simple attention mechanism for temporal predictions with transformers.APL Computational Physics, 1(1):016104, 2025

    2025

  46. [46]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need, 2023. URLhttps://arxiv.org/abs/1706.03762

    Pith/arXiv arXiv 2023

  47. [47]

    The minimal flow unit in near-wall turbulence.Journal of Fluid Mechanics, 225: 213–240, 1991

    Javier Jiménez and Parviz Moin. The minimal flow unit in near-wall turbulence.Journal of Fluid Mechanics, 225: 213–240, 1991

    1991

  48. [48]

    Eddies, streams, and convergence zones in turbulent flows

    Julian CR Hunt, Alan A Wray, and Parviz Moin. Eddies, streams, and convergence zones in turbulent flows. Studying turbulence using numerical simulation databases, 2. Proceedings of the 1988 summer program, 1988

    1988

  49. [49]

    Sod2d: A gpu-enabled spectral finite elements method for compressible scale-resolving simulations.Computer Physics Communications, 297:109067, 2024

    Lucas Gasparino, Filippo Spiga, and Oriol Lehmkuhl. Sod2d: A gpu-enabled spectral finite elements method for compressible scale-resolving simulations.Computer Physics Communications, 297:109067, 2024

    2024

  50. [50]

    Turbulence statistics in fully developed channel flow at low reynolds number.Journal of fluid mechanics, 177:133–166, 1987

    John Kim, Parviz Moin, and Robert Moser. Turbulence statistics in fully developed channel flow at low reynolds number.Journal of fluid mechanics, 177:133–166, 1987

    1987

  51. [51]

    Spatiotemporal analysis of complex signals: theory and applications.Journal of Statistical Physics, 64(3):683–739, 1991

    Nadine Aubry, Régis Guyonnet, and Ricardo Lima. Spatiotemporal analysis of complex signals: theory and applications.Journal of Statistical Physics, 64(3):683–739, 1991

    1991

  52. [52]

    Deep generative models for distribution-preserving lossy compression.Advances in neural information processing systems, 31, 2018

    Michael Tschannen, Eirikur Agustsson, and Mario Lucic. Deep generative models for distribution-preserving lossy compression.Advances in neural information processing systems, 31, 2018

    2018

  53. [53]

    Generative adversarial networks for extreme learned image compression

    Eirikur Agustsson, Michael Tschannen, Fabian Mentzer, Radu Timofte, and Luc Van Gool. Generative adversarial networks for extreme learned image compression. InProceedings of the IEEE/CVF International Conference on Computer Vision, pages 221–231, 2019

    2019

  54. [54]

    High-fidelity generative image compression.Advances in Neural Information Processing Systems, 33:11913–11924, 2020

    Fabian Mentzer, George D Toderici, Michael Tschannen, and Eirikur Agustsson. High-fidelity generative image compression.Advances in Neural Information Processing Systems, 33:11913–11924, 2020

    2020

  55. [55]

    Auto-encoding variational bayes.arXiv preprint arXiv:1312.6114, 2013

    Diederik P Kingma and Max Welling. Auto-encoding variational bayes.arXiv preprint arXiv:1312.6114, 2013

    Pith/arXiv arXiv 2013

  56. [56]

    Understanding disentangling inβ-vae.arXiv preprint arXiv:1804.03599, 2018

    Christopher P Burgess, Irina Higgins, Arka Pal, Loic Matthey, Nick Watters, Guillaume Desjardins, and Alexander Lerchner. Understanding disentangling inβ-vae.arXiv preprint arXiv:1804.03599, 2018

    Pith/arXiv arXiv 2018

  57. [57]

    Disentangling generative factors of physical fields using variational autoencoders.Frontiers in Physics, 10:890910, 2022

    Christian Jacobsen and Karthik Duraisamy. Disentangling generative factors of physical fields using variational autoencoders.Frontiers in Physics, 10:890910, 2022

    2022

  58. [58]

    Deconvolution and checkerboard artifacts.Distill, 2016

    Augustus Odena, Vincent Dumoulin, and Chris Olah. Deconvolution and checkerboard artifacts.Distill, 2016. doi:10.23915/distill.00003. URLhttp://distill.pub/2016/deconv-checkerboard

    work page doi:10.23915/distill.00003 2016

  59. [59]

    Spectral normalization for generative adversarial networks.arXiv preprint arXiv:1802.05957, 2018

    Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative adversarial networks.arXiv preprint arXiv:1802.05957, 2018. 26 A Hybrid Generative Reduced-Order ModelA PREPRINT

    Pith/arXiv arXiv 2018

  60. [60]

    Image-to-image translation with conditional adversarial networks

    Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translation with conditional adversarial networks. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 1125–1134, 2017

    2017

  61. [61]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Pith/arXiv arXiv 2014

  62. [62]

    Cyclical annealing schedule: A simple approach to mitigating kl vanishing.arXiv preprint arXiv:1903.10145, 2019

    Hao Fu, Chunyuan Li, Xiaodong Liu, Jianfeng Gao, Asli Celikyilmaz, and Lawrence Carin. Cyclical annealing schedule: A simple approach to mitigating kl vanishing.arXiv preprint arXiv:1903.10145, 2019

    Pith/arXiv arXiv 1903

  63. [63]

    On deep-learning-based closures for algebraic surrogate models of turbulent flows.Journal of Fluid Mechanics, 1020:A36, 2025

    Benet Eiximeno, Marcial Sanchis-Agudo, Arnau Miró, Ivette Rodriguez, Ricardo Vinuesa, and Oriol Lehmkuhl. On deep-learning-based closures for algebraic surrogate models of turbulent flows.Journal of Fluid Mechanics, 1020:A36, 2025. doi:10.1017/jfm.2025.10610

    work page doi:10.1017/jfm.2025.10610 2025

  64. [64]

    Accurate, large minibatch sgd: Training imagenet in 1 hour.arXiv preprint arXiv:1706.02677, 2017

    Priya Goyal, Piotr Dollár, Ross Girshick, Pieter Noordhuis, Lukasz Wesolowski, Aapo Kyrola, Andrew Tulloch, Yangqing Jia, and Kaiming He. Accurate, large minibatch sgd: Training imagenet in 1 hour.arXiv preprint arXiv:1706.02677, 2017

    Pith/arXiv arXiv 2017

  65. [65]

    Scheduled sampling for sequence prediction with recurrent neural networks

    Samy Bengio, Oriol Vinyals, Navdeep Jaitly, and Noam Shazeer. Scheduled sampling for sequence prediction with recurrent neural networks. In C. Cortes, N. Lawrence, D. Lee, M. Sugiyama, and R. Garnett, editors,Advances in Neural Information Processing Systems (NIPS), volume 28. Curran Associates, Inc., 2015

    2015

  66. [66]

    Importance weighted autoencoders.arXiv preprint arXiv:1509.00519, 2015

    Yuri Burda, Roger Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders.arXiv preprint arXiv:1509.00519, 2015

    Pith/arXiv arXiv 2015

  67. [67]

    Quadrant analysis in turbulence research: history and evolution.Annual Review of Fluid Mechanics, 48(1):131–158, 2016

    James M Wallace. Quadrant analysis in turbulence research: history and evolution.Annual Review of Fluid Mechanics, 48(1):131–158, 2016

    2016

  68. [68]

    Structure of the reynolds stress near the wall.Journal of Fluid Mechanics, 55(1): 65–92, 1972

    WW Willmarth and SS Lu. Structure of the reynolds stress near the wall.Journal of Fluid Mechanics, 55(1): 65–92, 1972. 27

    1972