pith. sign in

arxiv: 2606.22084 · v1 · pith:2ASPGHJHnew · submitted 2026-06-20 · ⚛️ physics.flu-dyn · cs.LG

Patched Flow Matching: Generative Wall-Pressure Reconstruction Beyond Training-Domain Scales from Sparse Sensors

Meet Hemant Parikh , Yi Liu , Jian-Xun Wang This is my paper

Pith reviewed 2026-06-26 11:29 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn cs.LG
keywords wall-pressure reconstructionflow matchinggenerative modelingsparse sensorsinner scalingReynolds-number transferturbulent channel flowpatch decomposition
0
0 comments X

The pith

Patched Flow Matching reconstructs full-resolution wall-pressure fields on domains four times larger than the training domain from sensor coverage as low as 0.25%.

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

The paper aims to establish that a generative model using patched flow matching can combine local priors learned from short-domain direct numerical simulations with sparse sensor measurements to produce complete wall-pressure fields over much larger areas. Short simulations capture small-scale high-wavenumber details but miss large-scale low-wavenumber patterns, while sensors supply the opposite information. By decomposing the problem into patches and expressing the prior in inner-scaled coordinates, the method decouples the learned statistics from overall domain size and Reynolds number. If correct, this supplies a route to the full wall-pressure spectrum that neither limited simulations nor sparse experiments can deliver on their own.

Core claim

Patched Flow Matching learns a patch-local prior over inner-scaled wall-pressure statistics from short-domain DNS and assimilates sparse sensor measurements at inference time through training-free posterior sampling. The patch-additive decomposition of the flow matching vector field decouples the generative prior from the global domain size, enabling reconstruction on domains four times larger than the training configuration from sensor coverage as low as 0.25%. By expressing the patch prior in inner-scaled coordinates, where high-wavenumber statistics are approximately Reynolds-number invariant, the framework extends to higher Reynolds numbers through hierarchical transfer learning with as

What carries the argument

The patch-additive decomposition of the flow matching vector field, which decouples the generative prior from the global domain size.

If this is right

  • Reconstruction of full-resolution wall-pressure fields on domains arbitrarily larger than the training configuration.
  • Recovery of low-wavenumber spectral content with high fidelity in both streamwise and spanwise directions.
  • Data-efficient extension to higher Reynolds numbers using only 500 short-domain snapshots, or 2.5% of the base training data.
  • Zero-shot generalization to unseen Reynolds numbers without retraining the patch prior.

Where Pith is reading between the lines

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

  • The same patch-decoupling strategy could be tested on other surface quantities such as shear stress if inner scaling continues to hold.
  • Integration with real experimental sensor arrays would allow reconstruction on facility-scale domains where long-domain DNS remains impossible.
  • The approach suggests that similar patch-based generative priors might reduce the cost of generating statistically stationary fields for even larger domain multiples.

Load-bearing premise

High-wavenumber wall-pressure statistics are approximately Reynolds-number invariant when expressed in inner-scaled coordinates.

What would settle it

A direct comparison in which the low-wavenumber portion of the reconstructed wall-pressure spectrum deviates substantially from independent long-domain DNS or high-resolution experiments at the same Reynolds number.

Figures

Figures reproduced from arXiv: 2606.22084 by Jian-Xun Wang, Meet Hemant Parikh, Yi Liu.

Figure 1
Figure 1. Figure 1: Wall-pressure fluctuations at Reτ = 180, 500, and 1000 visualized in (a) inner-scaled coordinates (x +, z+) and (b) outer-scaled coordinates (x/δ, z/δ). In inner units, windows of fixed viscous-unit size capture statistically similar high-wavenumber structures across Reynolds numbers, whereas the low-wavenumber content varies system￾atically with Reτ in both coordinate systems. This scale separation motiva… view at source ↗
Figure 2
Figure 2. Figure 2: Schematic of the Patched Flow Matching framework. (a) A snapshot of resolution [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Unconditional inference procedure of Patched FM. At each ODE step, a tiling origin [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Training-free conditional generation. At each ODE step, the Tweedie estimator produces a one-step pre [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Unconditional generation at Reτ = 180: (a) Contours of the wall-pressure fluctuations from DNS and two independently generated samples on the short training domain (L S x = 4πδ); (b) streamwise and spanwise wavenumber spectra Φpp(kx) and Φpp(kz) from an ensemble of 500 generated samples, compared with short-domain DNS. Conditional reconstruction on the long domain. We now verify one of the central hypothes… view at source ↗
Figure 6
Figure 6. Figure 6: Conditional reconstruction at Reτ = 180 on the long domain (L L x = 16πδ, Case 4): (a) Contours of the wall-pressure fluctuations from sparse sensor measurements (1% coverage), cubic interpolation, Patched FM reconstruction, and long-domain DNS ground truth; (b) streamwise and spanwise wavenumber spectra Φpp(kx) and Φpp(kz), comparing Patched FM, cubic interpolation, short-domain DNS, and long-domain DNS. … view at source ↗
Figure 7
Figure 7. Figure 7: Unconditional generation at Reτ = 500 after hierarchical transfer from Reτ = 180: (a) Contours of the wall￾pressure fluctuations from DNS and two generated samples on the short training domain (L S x = 4πδ); (b) streamwise and spanwise wavenumber spectra Φpp(kx) and Φpp(kz) from 100 generated samples, compared with DNS. For conditional reconstruction on the long domain (Case 5, 0.36% sensor coverage), [PI… view at source ↗
Figure 8
Figure 8. Figure 8: Conditional reconstruction at Reτ = 500 on the long domain (L L x = 16πδ, Case 5): (a) Contours of the wall-pressure fluctuations from sparse sensor measurements (0.36% coverage), cubic interpolation, Patched FM recon￾struction, and long-domain DNS; (b) streamwise and spanwise wavenumber spectra Φpp(kx) and Φpp(kz), comparing Patched FM, cubic interpolation, short-domain DNS, and long-domain DNS. increases… view at source ↗
Figure 9
Figure 9. Figure 9: Unconditional generation at Reτ = 1000 after hierarchical transfer from Reτ = 500: (a) Contours of the wall-pressure fluctuations from DNS and two generated samples on the short training domain (L S x = 4πδ); (b) streamwise and spanwise wavenumber spectra Φpp(kx) and Φpp(kz) from 100 generated samples, compared with DNS. tional interpolation for reconstructing turbulent wall-pressure fields from sparse mea… view at source ↗
Figure 10
Figure 10. Figure 10: Conditional reconstruction at Reτ = 1000 on the long domain (L L x = 16πδ, Case 6): (a) Contours of the wall-pressure fluctuations from sparse sensor measurements (0.25% coverage), cubic interpolation, Patched FM reconstruction, and long-domain DNS; (b) streamwise and spanwise wavenumber spectra Φpp(kx) and Φpp(kz), comparing Patched FM, cubic interpolation, short-domain DNS, and long-domain DNS. 4.1. Nec… view at source ↗
Figure 11
Figure 11. Figure 11: Comparison of Patched FM against a baseline flow matching model trained on full-resolution data at [PITH_FULL_IMAGE:figures/full_fig_p027_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Comparison of inner-scaled versus outer-scaled Patched FM models at [PITH_FULL_IMAGE:figures/full_fig_p029_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Generation performance under severely limited training data ( [PITH_FULL_IMAGE:figures/full_fig_p030_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Zero-shot conditional reconstruction at Reτ = 500 without any fine-tuning at the target Reynolds number: (a) Contours of the wall-pressure fluctuations from sparse sensor measurements (0.36% coverage), zero-shot recon￾structions using models trained at Reτ = 180 and Reτ = 1000, and long-domain DNS ground truth (L L x = 16πδ, Case 5); (b) streamwise (top) and spanwise (bottom) wavenumber spectra Φpp(kx) an… view at source ↗
Figure 15
Figure 15. Figure 15: Computational cost of the Patched FM framework. (a) Cumulative training time (minutes, single NVIDIA [PITH_FULL_IMAGE:figures/full_fig_p034_15.png] view at source ↗
read the original abstract

Characterizing the complete wall-pressure spectrum in turbulent wall-bounded flows requires simultaneous access to the viscous-scale high-wavenumber content and the outer-layer low-wavenumber content -- a requirement that neither short-domain direct numerical simulation (DNS) nor sparse experimental measurements alone can satisfy. We propose Patched Flow Matching (Patched FM), a generative framework that fuses these two complementary sources by learning a patch-local prior over inner-scaled wall-pressure statistics from short-domain DNS and assimilating sparse sensor measurements at inference time through training-free posterior sampling. The patch-additive decomposition of the flow matching vector field decouples the generative prior from the global domain size, enabling reconstruction on domains arbitrarily larger than the training configuration. By expressing the patch prior in inner-scaled coordinates, where high-wavenumber wall-pressure statistics are approximately Reynolds-number invariant, the framework extends to higher Reynolds numbers through hierarchical transfer learning with as few as $500$ short-domain snapshots ($2.5\%$ of the base training data) at a fraction of the scratch-training cost. Applied to compressible channel-flow DNS at $Re_\tau = 180$, $500$, and $1000$, Patched FM reconstructs full-resolution wall-pressure fields on a domain four times larger than the training configuration ($L_x^L = 16\pi\delta$ versus $L_x^S = 4\pi\delta$) from sensor coverage as low as $0.25\%$, recovering the low-wavenumber spectral content inaccessible to short-domain DNS with high fidelity in both streamwise and spanwise directions. Zero-shot generalization to unseen Reynolds numbers and ablation studies further confirm the role of inner scaling as a physical prerequisite for data-efficient Reynolds-number transfer.

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 / 1 minor

Summary. The manuscript introduces Patched Flow Matching (Patched FM), a generative framework for reconstructing full-resolution wall-pressure fields in turbulent wall-bounded flows. It learns a patch-local prior over inner-scaled high-wavenumber statistics from short-domain DNS (Lx^S = 4πδ) and assimilates sparse sensor measurements at inference via training-free posterior sampling. The patch-additive decomposition of the flow-matching vector field decouples the prior from global domain size, enabling zero-shot reconstruction on domains four times larger (Lx^L = 16πδ) from sensor coverages as low as 0.25%. By exploiting approximate Reynolds-number invariance of high-wavenumber wall-pressure statistics in inner units, the method performs hierarchical transfer learning to higher Re_τ with only 500 snapshots (2.5% of base data). Results are demonstrated on compressible channel-flow DNS at Re_τ = 180, 500, and 1000, with claims of high-fidelity recovery of low-wavenumber spectral content inaccessible to short-domain DNS and zero-shot generalization to unseen Reynolds numbers.

Significance. If the quantitative claims hold, the work would provide a practical route to obtaining complete wall-pressure spectra by fusing limited DNS priors with sparse measurements, addressing a long-standing limitation in turbulence research. The patch-additive decomposition is a clear technical strength that enables domain-size-independent generation. The use of inner scaling to justify data-efficient Re transfer is conceptually appealing and, if validated with explicit bounds, could reduce the cost of high-Re studies. No machine-checked proofs or open reproducible code are mentioned, but the framework's emphasis on physical invariance as a prerequisite for transfer is a positive feature.

major comments (2)
  1. [Abstract] Abstract: The central claims of 'high-fidelity' reconstruction and recovery of low-wavenumber content rest on quantitative performance, yet the abstract supplies no error metrics, spectral comparisons, error bars, or benchmark validations. This absence prevents evaluation of whether the reported reconstructions actually achieve the stated fidelity on the Lx^L = 16πδ domain.
  2. [Abstract] Abstract (hierarchical transfer learning paragraph): The data-efficiency claim (500 snapshots, 2.5% of base data) for Re_τ extension depends on the approximate invariance of high-wavenumber wall-pressure statistics in inner-scaled coordinates. No quantitative bound (e.g., integrated spectral difference for k^+ > 0.05 or Kolmogorov-Smirnov distance) or ablation isolating the effect of violating this assumption is provided; without it the transfer step and associated cost reduction cannot be assessed.
minor comments (1)
  1. [Abstract] Abstract: The phrase 'patch-additive decomposition of the flow matching vector field' is introduced without an accompanying equation or brief definition, which obscures the precise mechanism decoupling the prior from domain size.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback focused on the abstract. We agree that quantitative support is needed to substantiate the central claims and will revise the abstract accordingly while preserving its length constraints.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claims of 'high-fidelity' reconstruction and recovery of low-wavenumber content rest on quantitative performance, yet the abstract supplies no error metrics, spectral comparisons, error bars, or benchmark validations. This absence prevents evaluation of whether the reported reconstructions actually achieve the stated fidelity on the Lx^L = 16πδ domain.

    Authors: We agree that the abstract would be strengthened by explicit quantitative indicators. In the revised manuscript we will insert concise statements reporting the relative L2 reconstruction error, streamwise/spanwise spectral correlation coefficients, and associated variability for the Lx^L = 16πδ cases, drawn directly from the quantitative results already presented in Section 4. This change will allow readers to assess the claimed fidelity without leaving the abstract. revision: yes

  2. Referee: [Abstract] Abstract (hierarchical transfer learning paragraph): The data-efficiency claim (500 snapshots, 2.5% of base data) for Re_τ extension depends on the approximate invariance of high-wavenumber wall-pressure statistics in inner-scaled coordinates. No quantitative bound (e.g., integrated spectral difference for k^+ > 0.05 or Kolmogorov-Smirnov distance) or ablation isolating the effect of violating this assumption is provided; without it the transfer step and associated cost reduction cannot be assessed.

    Authors: The manuscript contains ablation studies on Reynolds-number transfer (Section 5) that demonstrate successful zero-shot generalization, but we acknowledge the abstract itself does not supply an explicit quantitative bound on the inner-scaled invariance. We will revise the abstract to include a brief statement of the observed spectral similarity (e.g., integrated difference for k^+ > 0.05) together with a reference to the ablation that isolates the assumption's role. This addresses the concern while remaining within abstract length limits. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on external DNS inputs and stated physical assumptions.

full rationale

The paper learns a patch-local prior from short-domain DNS snapshots and assimilates sparse sensor data at inference via training-free sampling. The inner-scaling invariance is invoked as an external physical property to justify hierarchical transfer with 500 snapshots, not derived from or defined by the model's own equations. No self-citations, fitted parameters renamed as predictions, or self-definitional reductions appear in the provided text. The central claims (domain extension, low-wavenumber recovery) are presented as outputs of the generative process conditioned on independent data, with ablations referenced to support the scaling choice. This satisfies the default expectation of a non-circular framework.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the physical assumption of inner-scale invariance of high-wavenumber wall-pressure statistics (a standard domain assumption in turbulence) and on the effectiveness of the generative prior learned from short-domain data; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption High-wavenumber wall-pressure statistics are approximately Reynolds-number invariant in inner-scaled coordinates.
    Invoked to justify hierarchical transfer learning with only 500 snapshots at higher Re.

pith-pipeline@v0.9.1-grok · 5851 in / 1332 out tokens · 30427 ms · 2026-06-26T11:29:33.524539+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

45 extracted references · 5 linked inside Pith

  1. [1]

    W. K. Blake, Mechanics of Flow-Induced Sound and Vibration: Complex Flow–Structure In- teractions, 2nd Edition, Vol. 2, Academic Press, 2017

    2017

  2. [2]

    Bull, Wall-pressure fluctuations beneath turbulent boundary layers: some reflections on forty years of research, Journal of Sound and Vibration 190 (3) (1996) 299–315

    M. Bull, Wall-pressure fluctuations beneath turbulent boundary layers: some reflections on forty years of research, Journal of Sound and Vibration 190 (3) (1996) 299–315

    1996

  3. [3]

    Devenport, N

    W. Devenport, N. Alexander, S. Glegg, M. Wang, The sound of flow over rigid walls, Annual Review of Fluid Mechanics 50 (2018) 435–458. 35

    2018

  4. [4]

    K. C. Kim, R. J. Adrian, Very large-scale motion in the outer layer, Physics of Fluids 11 (2) (1999) 417–422

    1999

  5. [5]

    Hutchins, I

    N. Hutchins, I. Marusic, Evidence of very long meandering features in the logarithmic region of turbulent boundary layers, Journal of Fluid Mechanics 579 (2007) 1–28

    2007

  6. [6]

    Marusic, R

    I. Marusic, R. Mathis, N. Hutchins, Predictive model for wall-bounded turbulent flow, Science 329 (5988) (2010) 193–196

    2010

  7. [7]

    Mathis, N

    R. Mathis, N. Hutchins, I. Marusic, Large-scale amplitude modulation of the small-scale struc- tures in turbulent boundary layers, Journal of Fluid Mechanics 628 (2009) 311–337

    2009

  8. [8]

    Y. Liu, K. Wang, M. Wang, Subconvective wall-pressure fluctuations in low-Mach-number turbulent channel flow, Journal of Fluid Mechanics 984 (2024) R2

    2024

  9. [9]

    Y. Liu, M. Wang, Reynolds number effect on wall-pressure fluctuations in low-Mach-number turbulent channel flow, AIAA Paper 2025–3492 (2025)

    2025

  10. [10]

    Lozano-Durán, J

    A. Lozano-Durán, J. Jiménez, Effect of the computational domain on direct simulations of turbulent channels up toReτ = 4200, Physics of Fluids 26 (1) (2014)

    2014

  11. [11]

    Jiménez, P

    J. Jiménez, P. Moin, The minimal flow unit in near-wall turbulence, Journal of Fluid Mechanics 225 (1991) 213–240

    1991

  12. [12]

    T.M.Farabee, M.J.Casarella, Spectralfeaturesofwallpressurefluctuationsbeneathturbulent boundary layers, Physics of Fluids A: Fluid Dynamics 3 (10) (1991) 2410–2420

    1991

  13. [13]

    Klewicki, P

    J. Klewicki, P. Priyadarshana, M. Metzger, Statistical structure of the fluctuating wall pressure and its in-plane gradients at high Reynolds number, Journal of Fluid Mechanics 609 (2008) 195–220

    2008

  14. [14]

    Gravante, A

    S. Gravante, A. Naguib, C. Wark, H. Nagib, Characterization of the pressure fluctuations under a fully developed turbulent boundary layer, AIAA journal 36 (10) (1998) 1808–1816

    1998

  15. [15]

    Hu, Sensor-size-related attenuation correction of wall pressure spectra measurements, Physics of Fluids 34 (6) (2022)

    N. Hu, Sensor-size-related attenuation correction of wall pressure spectra measurements, Physics of Fluids 34 (6) (2022). 36

    2022

  16. [16]

    S. L. Prigent, É. Salze, C. Bailly, Deconvolution of wave-number-frequency spectra of wall pressure fluctuations, AIAA Journal 58 (1) (2020) 164–173

    2020

  17. [17]

    Fukami, K

    K. Fukami, K. Fukagata, K. Taira, Super-resolution reconstruction of turbulent flows with machine learning, Journal of Fluid Mechanics 870 (2019) 106–120

    2019

  18. [18]

    Fukami, K

    K. Fukami, K. Fukagata, K. Taira, Machine-learning-based spatio-temporal super resolution reconstruction of turbulent flows, Journal of Fluid Mechanics 909 (2021) A9

    2021

  19. [19]

    N. K. Chaurasia, S. Chakraborty, Reconstruction of the turbulent flow field with sparse mea- surements using physics-informed neural network, Physics of Fluids 36 (8) (2024)

    2024

  20. [20]

    Guastoni, A

    L. Guastoni, A. Güemes, A. Ianiro, S. Discetti, P. Schlatter, H. Azizpour, R. Vinuesa, Convolutional-networkmodelstopredictwall-boundedturbulencefromwallquantities, Journal of Fluid Mechanics 928 (2021) A27

    2021

  21. [21]

    Balasubramanian, L

    A. Balasubramanian, L. Guastoni, A. Güemes, A. Ianiro, S. Discetti, P. Schlatter, H. Az- izpour, R. Vinuesa, Predicting the near-wall region of turbulence through convolutional neural networks, arXiv preprint arXiv:2107.07340 (2021)

    arXiv 2021

  22. [22]

    Y. Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, B. Poole, Score-based gen- erative modeling through stochastic differential equations, arXiv preprint arXiv:2011.13456 (2020)

    Pith/arXiv arXiv 2011

  23. [23]

    Lipman, R

    Y. Lipman, R. T. Chen, H. Ben-Hamu, M. Nickel, M. Le, Flow matching for generative mod- eling, arXiv preprint arXiv:2210.02747 (2022)

    Pith/arXiv arXiv 2022

  24. [24]

    Albergo, N

    M. Albergo, N. M. Boffi, E. Vanden-Eijnden, Stochastic interpolants: A unifying framework for flows and diffusions, Journal of Machine Learning Research 26 (209) (2025) 1–80

    2025

  25. [25]

    Chung, J

    H. Chung, J. Kim, M. T. Mccann, M. L. Klasky, J. C. Ye, Diffusion posterior sampling for general noisy inverse problems, arXiv preprint arXiv:2209.14687 (2022)

    Pith/arXiv arXiv 2022

  26. [26]

    J. Song, A. Vahdat, M. Mardani, J. Kautz, Pseudoinverse-guided diffusion models for inverse problems, in: International conference on learning representations, 2023. 37

    2023

  27. [27]

    M. H. Parikh, Y. Chen, J.-X. Wang, D-flow sgld: Source-space posterior sampling for scientific inverse problems with flow matching, arXiv preprint arXiv:2602.21469 (2026)

    arXiv 2026

  28. [28]

    H. Gao, X. Han, X. Fan, L. Sun, L.-P. Liu, L. Duan, J.-X. Wang, Bayesian conditional diffu- sion models for versatile spatiotemporal turbulence generation, Computer Methods in Applied Mechanics and Engineering 427 (2024) 117023

    2024

  29. [29]

    P. Du, M. H. Parikh, X. Fan, X.-Y. Liu, J.-X. Wang, Conditional neural field latent diffusion model for generating spatiotemporal turbulence, Nature Communications 15 (1) (2024) 10416

    2024

  30. [30]

    X. Fan, D. Akhare, J.-X. Wang, Neural differentiable modeling with diffusion-based super- resolution for two-dimensional spatiotemporal turbulence, Computer Methods in Applied Me- chanics and Engineering 433 (2025) 117478

    2025

  31. [31]

    Chakraborty, H

    D. Chakraborty, H. Kim, R. Maulik, Adaptive diffusion posterior sampling for data and model fusion of complex nonlinear dynamical systems, arXiv preprint arXiv:2603.12635 (2026)

    arXiv 2026

  32. [32]

    Vishwasrao, S

    A. Vishwasrao, S. B. C. Gutha, A. Cremades, K. Wijk, A. Patil, C. Gorle, B. J. McKeon, H. Azizpour, R. Vinuesa, Diff-SPORT: Diffusion-based sensor placement optimization and re- construction of turbulent flows in urban environments, arXiv preprint arXiv:2506.00214 (2025)

    arXiv 2025

  33. [33]

    X.-Y. Liu, M. H. Parikh, X. Fan, P. Du, Q. Wang, Y.-F. Chen, J.-X. Wang, CoNFiLD-inlet: Synthetic turbulence inflow using generative latent diffusion models with neural fields, Physical Review Fluids 10 (5) (2025) 054901

    2025

  34. [34]

    M. H. Parikh, X. Fan, J.-X. Wang, Conditional flow matching for generative modelling of near-wall turbulence with quantified uncertainty, Journal of Fluid Mechanics 1029 (2026) A32

    2026

  35. [35]

    X. Fan, M. H. Parikh, Y. Liu, X.-Y. Liu, J. Guo, M. Wang, J.-X. Wang, Generative reconstruc- tion of spatiotemporal wall-pressure in turbulent boundary layers via patchwise latent diffusion, arXiv preprint arXiv:2511.12455 (2025)

    arXiv 2025

  36. [36]

    B. E. McGrath, R. L. Simpson, Some features of surface pressure fluctuations in turbulent boundary layers with zero and favorable pressure gradients, Tech. Rep. NASA Contractor Report 4051, NASA (1987). 38

    1987

  37. [37]

    H. Choi, P. Moin, On the space-time characteristics of wall-pressure fluctuations, Physics of Fluids A: Fluid Dynamics 2 (8) (1990) 1450–1460

    1990

  38. [38]

    B. Yang, Z. Yang, On the wavenumber–frequency spectrum of the wall pressure fluctuations in turbulent channel flow, Journal of Fluid Mechanics 937 (2022) A39

    2022

  39. [39]

    Gloerfelt, Aerodynamic noise from turbulent channel flow using compressible simulations, AIAA Paper 2024–3074 (2024)

    X. Gloerfelt, Aerodynamic noise from turbulent channel flow using compressible simulations, AIAA Paper 2024–3074 (2024)

    2024

  40. [40]

    J. Hu, B. Song, X. Xu, L. Shen, J. A. Fessler, Learning image priors through patch-based diffu- sion models for solving inverse problems, Advances in Neural Information Processing Systems 37 (2024) 1625–1660

    2024

  41. [41]

    A. J. Smits, B. J. McKeon, I. Marusic, High–Reynolds number wall turbulence, Annual Review of Fluid Mechanics 43 (1) (2011) 353–375

    2011

  42. [42]

    A. Tong, K. Fatras, N. Malkin, G. Huguet, Y. Zhang, J. Rector-Brooks, G. Wolf, Y. Bengio, Improving and generalizing flow-based generative models with minibatch optimal transport, arXiv preprint arXiv:2302.00482 (2023)

    Pith/arXiv arXiv 2023

  43. [43]

    Y. Xu, S. Luo, L. Wang, D. He, C. Liu, Diagnosing and improving diffusion models by esti- mating the optimal loss value, arXiv preprint arXiv:2506.13763 (2025)

    Pith/arXiv arXiv 2025

  44. [44]

    Bonnaire, R

    T. Bonnaire, R. Urfin, G. Biroli, M. Mézard, Why diffusion models don’t memorize: The role of implicit dynamical regularization in training, Advances in Neural Information Processing Systems 38 (2026) 141266–141286

    2026

  45. [45]

    Buchanan, D

    S. Buchanan, D. Pai, Y. Ma, V. De Bortoli, On the edge of memorization in diffusion models, Advances in Neural Information Processing Systems 38 (2026) 96113–96157. Appendix A. Supplementary Studies Appendix A.1. Effect of number of sensors on generation performance The conditional generation framework relies on sparse sensor observations to guide the rec...

    2026