Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers

Changhoon Lee; Junhyuk Kim

REVIEW 3 major objections 6 minor 40 references

Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers

T0 review · 3 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a GAN trained on turbulence data at three Reynolds numbers, combined with an RNN that advances its latent state in time, can generate instantaneous and time-varying channel-flow inflow fields at trained and…

desk verdict A solid, workmanlike GAN-plus-RNN inflow generator whose strongest result is long-time stationary generation at trained Reynolds numbers; the untrained-Reynolds interpolation claim is the soft spot and needs sharper framing. read the letter →

arxiv 1908.10515 v1 pith:OI7ZL2VA submitted 2019-08-28 physics.flu-dyn physics.comp-ph

Junhyuk Kim , Changhoon Lee This is my paper · ORCID

classification physics.flu-dynphysics.comp-ph

keywords turbulentinflowgenerationgenerativeadversarialnetworksunsupervisedlearningrecurrentneuralchannelflowReynoldsnumberinterpolationdirectnumericalsimulationsyntheticturbulence

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Turbulent inflow generation for channel-flow simulations normally requires either an expensive auxiliary direct numerical simulation (DNS) or hand-tuned synthetic eddy methods. This paper claims that one unsupervised generative model can replace both for a range of Reynolds numbers: a generative adversarial network trained only on two-dimensional cross-sectional velocity and temperature fields at three friction Reynolds numbers learns a latent direction that encodes the Reynolds-number effect, so it can generate realistic instantaneous fields at trained and untrained Reynolds numbers. When that generator is combined with an independently recurrent LSTM that advances the latent vector in time, the model produces statistically stationary, time-varying fields for at least 10,000 steps, with spanwise energy spectra and temporal correlations in good agreement with DNS. The payoff, if the claim holds, is cheap, parameter-flexible inflow conditions and evidence that turbulence statistics can be learned as a distribution rather than predicted deterministically.

What carries the argument

The load-bearing mechanism is the GAN latent vector $\mathbf{z}$, treated as a low-dimensional semantic code. One component $z_1$ is given the meaning of friction Reynolds number by appending a constant 'Re map' channel to the discriminator input; the WGAN-GP loss with gradient penalty and an energy-spectrum statistical constraint ($\mathrm{MSE_G}$) trains the generator to reproduce DNS two-point statistics. Time dependence is carried by an IndyLSTM—a recurrent network whose recurrent weights act elementwise rather than through a full matrix—which updates $\mathbf{z}_t$ to $\mathbf{z}_{t+1}$ from white noise; two auxiliary discriminators, $D_{\mathrm{time}}$ on short field sequences and $D_{\mathrm{norm}}$ on the latent outputs' distribution, keep the generated sequence statistically stationary and decorrelated over long horizons.

What would settle it

Run the trained RNN-GAN at an untrained Reynolds number such as $\mathrm{Re}_\tau=270$ as the actual inlet condition of a developing channel-flow DNS and measure skin-friction coefficient and streamwise spectra at $x^+ \approx 1000$ downstream. If the flow does not match a DNS with a faithful inlet, or if the inlet fields lose high-wavenumber energy within a few eddy-turnover times, the inflow-generation claim fails.

Watch

Extended reading notes

Core claim

The central discovery is that the latent space of a WGAN-GP trained on y–z plane fields at $\mathrm{Re}_\tau = 180$, $360$, and $540$ contains a semantically meaningful Reynolds-number direction. Fixing one latent component to a target value and feeding a constant Re map to the discriminator lets the generator produce fields whose mean velocity, rms fluctuation, and vorticity statistics follow DNS at both trained and untrained Reynolds numbers—$270$, $450$, and $720$—with accuracy that is clearly better inside the trained interval than outside it. The same generator, driven by an independently recurrent LSTM (IndyLSTM) that consumes random noise and is constrained by temporal- and normality-discriminators, yields a long statistically stationary sequence whose spanwise spectra and temporal correlations match DNS, in contrast to an autoencoder baseline that blurs small-scale structure. The paper concludes that an RNN-GAN trained this way can serve as a synthetic inflow generator for turbulent channel flow.

Load-bearing premise

The central fragile premise is that linearly varying the Reynolds-number component of the latent vector, guided only by a constant label shown to the discriminator, produces physically correct intermediate fields; nothing guarantees this smoothness, and the paper's evidence for it is empirical.

Editorial extensions

If this is right

Inside the trained interval ($180 \le \mathrm{Re}_\tau \le 540$), the model can supply inlet fields without an auxiliary DNS or a fully developed initial field, so the cost of inlet preparation drops to a single forward pass.
Because temporal correlations do not rebound as they do in the periodic DNS domain, training data do not require extremely long streamwise domains, which simplifies future data collection.
The spanwise width of the generated inlet can be enlarged by stitching independent latent maps, so one trained model can feed simulations wider than the training domain.
Beyond $\mathrm{Re}_\tau = 540$, statistics degrade, so practical use should stay within or near the trained range unless the model is extended by a physics-based loss.
The same latent-coding trick could carry other dimensionless parameters, such as Prandtl number, giving thermal inflow conditions at untrained values.

Reading between the lines

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

A decisive test the paper does not report is a posteriori: use the generated fields as the actual inlet of a developing channel-flow simulation and compare downstream skin friction, heat transfer, and spectra with a reference DNS; the paper only validates the inlet fields themselves.
If the latent-smoothness assumption generalizes, conditioning other latent components on parameters such as pressure gradient or wall temperature would give a single network as a cheap parameter sweep across flow conditions.
The reported training cost—about three weeks on one GPU for 17 million parameters—should be weighed against the alternative of running a new auxiliary DNS for each target Reynolds number; inference itself is cheap, so the model is most attractive when many distinct inflow conditions are needed.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Bottom line: this is a solid engineering contribution to inflow generation, not a fundamental advance in turbulence modeling. The strongest and most convincing part is the RNN-GAN's ability to produce statistically stationary, time-varying fields for 10,000 steps with spectral and temporal correlations close to DNS at a trained Reynolds number; the comparison against the autoencoder baseline is fair and useful. The novelty is real but modest: a GAN conditioned on Reynolds number through a z1 component and a constant Re map, plus an IndyLSTM operating in latent space for time advancement. The paper also documents sensible engineering choices—mirror and spectral augmentation, streamwise-vorticity input to the discriminator, statistical constraints—and it is honest about several limitations.

The soft spots are in the Reynolds-number interpolation claim. The abstract says the GAN can produce fields at various Re 'without any additional simulation,' based on training at three Re values. That claim rests on z1 behaving as a semantically meaningful, monotone Reynolds-number dial, and nothing in the WGAN-GP loss guarantees that; the discriminator only sees the three trained Re values. The evidence at Re_tau = 270, 450, 720 is qualitative—visual similarity and a few statistics—with no uncertainty quantification and no description of how the comparison DNS data at those Re were obtained. The authors admit extrapolation beyond Re_tau = 540 is inaccurate and that the Reynolds-number effect is 'not perfectly reflected.' That is the right instinct, and it means the paper's public claim needs softening more than its architecture needs fixing.

Second, the statistical constraints in Eqs. (7) and (24) directly penalize the same spanwise energy spectra and temporal correlations that are later shown as evidence in Figs. 10 and 11. So those figures are partly in-sample; they certify that the constrained loss was optimized, not independently that the GAN learned turbulence. The paper cites Wu et al. for this trick, so this is a known modeling choice, but the distinction between constrained and held-out metrics should be explicit. This is a real concern, though it does not undermine the core long-time-stationarity result, since that test is not itself constrained.

The citation pattern looks fine: relevant prior work on inflow generation, GANs, statistical constraints, and LSTM turbulence models is there. No code or data is released, which for a methods paper in 2019 is a limitation but not disqualifying.

Who should read it: anyone building or benchmarking synthetic inflow generators, especially with machine learning. It deserves peer review—it is a legitimate contribution—but I would send it back with a request to separate constrained metrics from holdouts and to frame the untrained-Re evidence as an empirical interpolation test rather than a demonstrated universal property. If the authors can release the trained model or code, that would materially raise its value.

Referee Report

3 major / 6 minor

Summary. The manuscript presents an RNN-GAN framework for generating turbulent channel-flow inlet fields in a y-z cross-section. A WGAN-GP generator/discriminator pair is trained on DNS data at Reτ = 180, 360, and 540, conditioned on Reynolds number through an additional "Re map" input channel and augmented by a spanwise-spectrum statistical constraint (Eq. 7). The trained generator is combined with an IndyLSTM RNN whose output latent vectors are decoded by the generator; additional discriminators Dtime and Dnorm and temporal-correlation/autocorrelation constraints (Eqs. 19–25) are used to obtain statistically stationary time series. The authors report good agreement with DNS for mean and rms profiles, spanwise spectra, and temporal correlations at trained Reynolds numbers, and similar (though degraded) agreement at untrained Reτ = 270, 450, and 720. They also demonstrate long-time stationarity over 10,000 steps, stochastic variation, and extension to larger spanwise domains, and compare against an autoencoder baseline.

Significance. If the central claim holds, the work offers a practical, fast synthetic inflow generator and an interesting demonstration that GANs can capture Reynolds-number-dependent near-wall turbulence statistics. The paper's strengths include extensive statistical comparisons at multiple Reynolds numbers, a long-time generation test exceeding the training horizon by a factor of about 20, a fair comparison against an autoencoder baseline, and physically motivated architectural choices (periodic padding, spectral augmentation, statistical constraints). However, the headline claim of generating at untrained Reynolds numbers rests on an unverified linear interpolation in latent space, and the reported spectral/temporal agreement is partly guaranteed by the loss functions. The extrapolation beyond the trained range is acknowledged to be inaccurate, and no uncertainty quantification is provided. The contribution is therefore promising but not yet fully substantiated.

major comments (3)

[§3.2, Eqs. (23)–(24) and Figs. 10–11] The spanwise energy spectrum and temporal autocorrelation are explicitly included as statistical constraints in the generator loss: S1 is the spectrum and S2 the temporal correlation in Eq. (24), with MSEG in Eq. (23) also based on the spectrum. The agreement shown in Figs. 10 and 11 is therefore partly a consequence of the training objective, not an independent validation of the model. Please state which validation metrics are not present in the loss and report them (for example, two-point spatial correlations, Reynolds-stress budgets, or probability density functions), or otherwise describe the degrees of freedom that make the spectral/temporal agreement nontrivial.
[§3.1 and §4 (Figs. 6, 7, 12, 13)] The claim that the model generates physically correct fields at Reynolds numbers not seen in training relies on the assumption that linear variation of the first latent component z1 interpolates meaningful turbulence statistics. The training loss does not constrain intermediate z1 values beyond the three trained Reynolds numbers, and GAN latent interpolation is an empirical heuristic. The paper's evidence at Reτ = 270, 450, and 720 is qualitative (Figs. 6 and 12) plus a few profile comparisons, and the paper itself acknowledges that extrapolation beyond Reτ = 540 is inaccurate. To support the central claim, please provide a quantitative holdout evaluation at untrained Reynolds numbers, including independent DNS validation data described in Section 2, error bars, and preferably a comparison against a baseline that simply interpolates the DNS statistics at the three training Reynolds numbers.
[§3.1, Fig. 7 and §3.2, Fig. 13] The statistical validation at untrained Reynolds numbers lacks uncertainty quantification. Figures 7 and 13 show profile agreements and visual similarity, but no confidence intervals, ensemble-size sensitivity, or quantitative error norms are reported; the described ensemble size of 10,000 generated fields would allow such estimates. Without them, phrases such as "very good agreement" and "fairly good agreement" are difficult to assess, especially where the generated profiles visibly deviate (e.g., Reτ = 720 in Fig. 7).

minor comments (6)

[§3.1] The word "deails" in the paragraph on data augmentation should be "details."
[§3.2] In the discussion of domain-size extension, "RAN-GAN" is a typo for "RNN-GAN."
[§3.1] The description of the Re map is ambiguous: the text says it is filled with the pixel value of the Reynolds number (z1 value) and also with wall coordinates containing Reynolds number information; specify exactly what the sixth input channel contains.
[§2] The DNS validation data at Reτ = 270, 450, and 720 are not described; state their provenance, grid resolution, sampling time, and averaging time so that the holdout comparisons are reproducible.
[§3.2, Eq. (25)] The moments S3 are not defined explicitly; list the formulas for mean, rms, skewness, and flatness used in the normal-distribution constraint.
[Fig. 7(b)] The inset figure should state which quantities are plotted and what the fitted log curve represents, so that the scatter of the maximum streamwise velocity fluctuation can be interpreted.

Circularity Check

1 steps flagged · score 6.0 of 10

The RNN-GAN's spanwise-spectrum and temporal-correlation validation is partly by construction: Eqs. (23)-(24) put exactly those statistics in the loss; the untrained-Reynolds-number claim remains an independent holdout test.

fitted input called prediction [Section 3.2, Eqs. (23)-(24), Figs. 10-11; also Section 3.1 Eq. (7)]
"MSEG =λ1‖S (X)− S ( ˜X)‖2 (7) ... MSERNN-G =λ1‖S 1(Tt)− S 1( ˜Tt)‖2 +λ2‖S 2(Tt)− S 2( ˜Tt)‖2 (24) ... The spanwise energy spectrums near the wall and the center, which were time-averaged for 50≤ t≤ 20050 excluding the transient period (1≤ t < 50), are presented in Fig. 10. ... The temporal correlations of the velocity components and temperature of the ﬁeld generated by the RNN-GAN are very favorably compared with those of the DNS in Fig. 11."

The same statistics that the paper presents as successful validation (spanwise energy spectrum S1 and temporal correlations S2 in Figs. 10 and 11) are directly minimized in the training objective. Eq. (7) penalizes ||S(X)-S(X̃)||2 for the instantaneous generator, and Eq. (24) penalizes ||S1(Tt)-S1(T̃t)||2 and ||S2(Tt)-S2(T̃t)||2 for the RNN-GAN time series. Therefore the agreement shown in Figs. 10 and 11 is in-sample by construction, not an independent test of the model. This does not make the entire paper circular: the mean and rms profiles, the pointwise correlations in Fig. 4, and the interpolation to untrained Reynolds numbers Reτ = 270, 450, 720 are not directly contained in the loss and provide independent support.

full rationale

No load-bearing self-citation is present: the cited GAN, WGAN-GP, Karras architecture, and statistical-constraint works are external, and the latent-space interpolation property is attributed to Radford/Berthelot rather than to the present authors. The various-Reynolds-number claim rests on a genuine holdout test at Reτ = 270, 450, and 720, with visual and statistical comparisons, so it is not circular; it is an empirical generalization of a conditional GAN with a Re map, not a derivation. The specific circularity is narrower: for the statistical quantities S (spanwise energy spectrum) and S2 (temporal correlations), the loss functions in Eqs. (7) and (23)-(24) directly force the generated values to approach the DNS values, and Figs. 10-11 then report that agreement as a result. Those validation panels are therefore partly by construction. Because the paper's headline claim of a successful synthetic inflow generator also relies on these spatiotemporal statistics, the circularity score is elevated to 6, but not higher, since the mean/rms statistics and the untrained-Reynolds-number behavior are genuinely out-of-sample empirical evidence.

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

The model is trained on DNS data and uses hand-tuned loss weights to enforce spectral and temporal statistics. The key assumption is that the GAN latent space is smooth in the Reynolds number direction, which is not derived from physics.

free parameters (4)

Lambda_1 (spectral constraint weight) = 10
Weights in MSEG Eq. (7); tuned to balance statistical constraint against GAN loss.
Lambda_2 (temporal correlation weight) = 1000
Weight in MSERNN-G Eq. (24); chosen by hand.
Lambda_3 (normal distribution moment weight) = 100
Weight in MSERNN Eq. (25).
Lambda_4 (autocorrelation weight) = 1
Weight for decorrelation guide in MSERNN Eq. (25).

assumptions (4)

domain assumption DNS data at three Reynolds numbers are representative of turbulent channel flow statistics.
Training data is ground truth; the GAN learns from these simulations.
ad hoc to paper The latent space of the trained GAN is semantically smooth, so linear interpolation in z1 yields physically meaningful variation in Reynolds number.
Relies on observed GAN latent space interpolation properties; not formally derived.
domain assumption Spanwise energy spectrum is a sufficient statistical constraint to ensure the generated fields are turbulent.
The statistical constraint uses only the one-dimensional spanwise spectrum, not full two-point correlations.
ad hoc to paper The RNN can learn a stationary stochastic process in the latent space that, when decoded by G, yields physically correct time series.
Assumed rather than proven; mitigated by Dnorm and long training sequences.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers." pith.science (2026). https://pith.science/paper/OI7ZL2VA

@misc{pith2026190810515,
  author       = {Pith},
  title        = {Pith review of: Deep unsupervised learning of turbulence for inflow generation at various Reynolds numbers},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OI7ZL2VA}},
  note         = {Machine review of arXiv:1908.10515}
}

read the original abstract

A realistic inflow boundary condition is essential for successful simulation of the developing turbulent boundary layer or channel flows. Recent advances in artificial intelligence (AI) have enabled the development of an inflow generator that performs better than the synthetic methods based on intuitions. In the present work, we applied generative adversarial networks (GANs), a representative of unsupervised learning, to generate an inlet boundary condition of turbulent channel flow. Upon learning the two-dimensional spatial structure of turbulence using data obtained from direct numerical simulation (DNS) of turbulent channel flow, the GAN could generate instantaneous flow fields that are statistically similar to those of DNS. Surprisingly, the GAN could produce fields at various Reynolds numbers without any additional simulation based on the trained data of only three Reynolds numbers. This indicates that the GAN could learn the universal nature of Reynolds number effect and might reflect other simulation conditions. Eventually, through a combination of the GAN and a recurrent neural network (RNN), we developed a novel model (RNN-GAN) that could generate time-varying fully developed flow for a long time. The spatiotemporal correlations of the generated flow are in good agreement with those of the DNS. This proves the usefulness of unsupervised learning in the generation of synthetic turbulence fields.

Figures

Figures reproduced from arXiv: 1908.10515 by the authors.

**Figure 2.** Architecture of (a) generator and (b) discriminator networks. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Generated fields at Reτ = 180. The streamwise vorticity is normalized by Reτ. In addition, the basic statistics of the generated fields, including the mean, root-mean-square (rms) of fluctuations, and pointwise correlation, are compared with those of the DNS in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Figures from the paper (12 more)

**Figure 4.** Figure 4: Statistics of fields generated by GAN at [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]

**Figure 5.** Figure 5: Collected data at three Reynolds numbers. (a) Streamwise velocity fields with grid index coordinates. (b) Streamwise velocity fields [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]

**Figure 6.** Figure 6: The generated fields by GAN at several Reynolds numbers. (a) and (b) are the streamwise velocity and vorticity fields normalized [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

**Figure 7.** Figure 7: Statistics of the fields generated by GAN at several Reynolds numbers. (a) Streamwise mean velocity with wall coordinates. (b) Mean [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

**Figure 8.** Figure 8: Schematic diagram of RNN-GAN. was used by Saito et al. [39]. The major difference is that we use the RNN to generate data for a very long period and do not fix a generated time-length. Our loss functions for training the model can be described as follows. Discriminator…

**Figure 9.** Figure 9: Time-dependently generated flow fields by RNN-GAN at [PITH_FULL_IMAGE:figures/full_fig_p016_9.png]

**Figure 10.** Figure 10: One-dimensional energy spectra results of RNN-GAN at [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]

**Figure 11.** Figure 11: Time correlations of generated flow by RNN-GAN at [PITH_FULL_IMAGE:figures/full_fig_p017_11.png]

**Figure 12.** Figure 12: Time-dependently generated streamwise velocity fields by RNN-GAN at various Reynolds numbers. (a,b) are at [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]

**Figure 13.** Figure 13: The rms profiles of (a) streamwise velocity, (b) vertical and spanwise velocity, (c) vorticity, and (d) temperature generated by [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]

**Figure 14.** Figure 14: Two flow fields generated by the RNN-GAN with di [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]

**Figure 15.** Figure 15: Extended fields in the spanwise direction at [PITH_FULL_IMAGE:figures/full_fig_p021_15.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 26 canonical work pages

[1]

LeCun, Y

Y . LeCun, Y . Bengio, G. Hinton, Deep learning, Nature 521 (2015) 436–444

2015
[2]

Goodfellow, J

I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, Y . Bengio, Generative adversarial nets, in: NIPS, 2014, pp. 2672–2680

work page 2014
[3]

Tabor, M

G. Tabor, M. Baba-Ahmadi, Inlet conditions for large eddy simulation: A review, Comput. Fluids 39 (2010) 553–567

work page 2010
[4]

Wu, Inﬂow turbulence generation methods, Annu

X. Wu, Inﬂow turbulence generation methods, Annu. Rev. Fluid Mech. 49 (2017) 23–49

work page 2017
[5]

J. Kim, P. Moin, R. Moser, Turbulence statistics in fully developed channel ﬂow at low reynolds number, J. Fluid Mech. 177 (1987) 133

work page 1987
[6]

P. R. Spalart, Direct simulation of a turbulent boundary layer up to Rθ = 1410, J. Fluid Mech. 187 (1988) 61–98

work page 1988
[7]

T. S. Lund, X. Wu, K. D. Squires, Generation of turbulent inﬂow data for spatially-developing boundary layer simulations, J. Comput. Phys. 140 (1998) 233–258

work page 1998
[8]

H. Le, P. Moin, J. Kim, Direct numerical simulation of turbulent ﬂow over a backward-facing step, J. Fluid Mech. 330 (1997) 349–374

work page 1997

Show all 40 references

[9]

Klein, A

M. Klein, A. Sadiki, J. Janicka, A digital ﬁlter based generation of inﬂow data for spatially developing direct numerical or large eddy simulations, J. Comput. Phys. 186 (2003) 652–665. 26 Kim and Lee / Journal of Computational Physics (2019)

2003
[10]

Jarrin, S

N. Jarrin, S. Benhamadouche, D. Laurence, R. Prosser, A synthetic-eddy-method for generating inﬂow conditions for large-eddy simulations, Int. J. Heat Fluid Flow 27 (2006) 585–593

2006
[11]

G. Oh, K. Noh, H. Park, J.-I. Choi, Extended synthetic eddy method to generate inﬂow data for turbulent thermal boundary layer, Int. J. Heat Mass Transf. 134 (2019) 1261–1267

2019
[12]

Keating, U

A. Keating, U. Piomelli, E. Balaras, H.-J. Kaltenbach, A priori and a posteriori tests of inﬂow conditions for large-eddy simulation, Phys. Fluids 16 (2004) 4696–4712

2004
[13]

R. King, O. Hennigh, A. Mohan, M. Chertkov, From deep to physics-informed learning of turbulence: Diagnostics, arXiv preprint arXiv:1810.07785v2 (2018)

2018 arXiv
[14]

Mohan, D

A. Mohan, D. Daniel, M. Chertkov, D. Livescu, Compressed convolutional LSTM: An e ﬃcient deep learning framework to model high ﬁdelity 3D turbulence, arXiv preprint arXiv:1903.00033v2 (2019)

2019 arXiv
[15]

Fukami, Y

K. Fukami, Y . Nabae, K. Kawai, K. Fukagata, Synthetic turbulent inﬂow generator using machine learning, Phys. Rev. Fluids 4 (2019)

2019
[16]

S. Lee, D. You, Prediction of laminar vortex shedding over a cylinder using deep learning, arXiv preprint arXiv:1712.07854v1 (2017)

2017 arXiv
[17]

S. Lee, D. You, Data-driven prediction of unsteady ﬂow ﬁelds over a circular cylinder using deep learning, arXiv preprint arXiv:1804.06076v2 (2018)

2018 arXiv
[18]

R ¨uttgers, S

M. R ¨uttgers, S. Lee, S. Jeon, D. You, Prediction of a typhoon track using a generative adversarial network and satellite images, Sci. Reports 9 (2019)

2019
[19]

Radford, L

A. Radford, L. Metz, S. Chintala, Unsupervised representation learning with deep convolutional generative adversarial networks, arXiv preprint arXiv:1511.06434v2 (2015)

2015 arXiv
[20]

Arjovsky, S

M. Arjovsky, S. Chintala, L. Bottou, Wasserstein GAN, arXiv preprint arXiv:1701.07875v3 (2017)

2017 arXiv
[21]

Gulrajani, F

I. Gulrajani, F. Ahmed, M. Arjovsky, V . Dumoulin, A. Courville, Improved training of Wasserstein GANs, in: NIPS, 2017, pp. 5767–5777

2017
[22]

Karras, T

T. Karras, T. Aila, S. Laine, J. Lehtinen, Progressive growing of GANs for improved quality, stability, and variation, arXiv preprint arXiv:1710.10196v3 (2017)

2017 arXiv
[23]

K. Roth, A. Lucchi, S. Nowozin, T. Hofmann, Stabilizing training of generative adversarial networks through regularization, in: In NIPS, 2017, pp. 2018–2028

2017
[24]

Mescheder, A

L. Mescheder, A. Geiger, S. Nowozin, Which training methods for GANs do actually converge?, arXiv preprint arXiv:1801.04406v4 (2018)

2018 arXiv
[25]

Karras, S

T. Karras, S. Laine, T. Aila, A style-based generator architecture for generative adversarial networks, arXiv preprint arXiv:1812.04948v3 (2018)

2018 arXiv
[26]

Y . Xie, E. Franz, M. Chu, N. Thuerey, tempoGAN: A temporally coherent, volumetric GAN for super-resolution ﬂuid ﬂow, arXiv preprint arXiv:1801.09710v2 (2018)

2018 arXiv
[27]

J.-L. Wu, K. Kashinath, A. Albert, D. Chirila, Prabhat, H. Xiao, Enforcing statistical constraints in generative adversarial networks for modeling chaotic dynamical systems, arXiv preprint arXiv:1905.06841v1 (2019)

2019 arXiv
[28]

R. D. Moser, J. Kim, N. N. Mansour, Direct numerical simulation of turbulent channel ﬂow up to Reτ=590, Phys. Fluids 11 (1999) 943–945

1999
[29]

Garc ´ıa-Villalba, J

M. Garc ´ıa-Villalba, J. C. del ´Alamo, Turbulence modiﬁcation by stable stratiﬁcation in channel ﬂow, Phys. Fluids 23 (2011) 045104

2011
[30]

D. P. Kingma, J. L. Ba, Adam: A method for stochastic optimization, arXiv preprint arXiv:1412.6980 (2014)

2014 arXiv
[31]

Abadi, A

M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis, J. Dean, M. Devin, S. Ghemawat, I. Good- fellow, A. Harp, G. Irving, M. Isard, Y . Jia, R. Jozefowicz, L. Kaiser, M. Kudlur, J. Levenberg, D. Man ´e, R. Monga, S. Moore, D. Mur- ray, C. Ola...

2015
[32]

Berthelot, T

D. Berthelot, T. Schumm, L. Metz, BEGAN: boundary equilibrium generative adversarial networks, arXiv preprint arXiv:1703.10717v4 (2017)

2017 arXiv
[33]

M. Lee, R. D. Moser, Direct numerical simulation of turbulent channel ﬂow up to Reτ=5200, J. Fluid Mech. 774 (2015) 395–415

2015
[34]

S. Li, W. Li, C. Cook, C. Zhu, Y . Gao, Independently recurrent neural network (indRNN): Building a longer and deeper RNN, arXiv preprint arXiv:1803.04831v3 (2018)

2018 arXiv
[35]

Gonnet, T

P. Gonnet, T. Deselaers, IndyLSTMs: Independently recurrent LSTMs, arXiv preprint arXiv:1903.08023v1 (2019)

2019 arXiv
[36]

Z. Wang, D. Xiao, F. Fang, R. Govindan, C. C. Pain, Y . Guo, Model identiﬁcation of reduced order ﬂuid dynamics systems using deep learning, Int. J. Numer. Methods Fluids 86 (2017) 255–268

2017
[37]

A. T. Mohan, D. V . Gaitonde, A deep learning based approach to reduced order modeling for turbulent ﬂow control using LSTM neural networks, arXiv preprint arXiv:1804.09269v1 (2018)

2018 arXiv
[38]

Cooijmans, N

T. Cooijmans, N. Ballas, C. Laurent, G ¨ulc ¸ehre, A. Courville, Recurrent batch normalization, arXiv preprint arXiv:1603.09025v5 (2016)

2016 arXiv
[39]

Saito, E

M. Saito, E. Matsumoto, S. Saito, Temporal generative adversarial nets with singular value clipping, in: ICCV , 2017

2017
[40]

Quadrio, P

M. Quadrio, P. Luchini, Integral space-time scales in turbulent wall ﬂows, Phys. Fluids 15 (2003) 2219–2227

2003

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

LeCun, Y

Y . LeCun, Y . Bengio, G. Hinton, Deep learning, Nature 521 (2015) 436–444

2015

[2] [2]

Goodfellow, J

I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, Y . Bengio, Generative adversarial nets, in: NIPS, 2014, pp. 2672–2680

work page 2014

[3] [3]

Tabor, M

G. Tabor, M. Baba-Ahmadi, Inlet conditions for large eddy simulation: A review, Comput. Fluids 39 (2010) 553–567

work page 2010

[4] [4]

Wu, Inﬂow turbulence generation methods, Annu

X. Wu, Inﬂow turbulence generation methods, Annu. Rev. Fluid Mech. 49 (2017) 23–49

work page 2017

[5] [5]

J. Kim, P. Moin, R. Moser, Turbulence statistics in fully developed channel ﬂow at low reynolds number, J. Fluid Mech. 177 (1987) 133

work page 1987

[6] [6]

P. R. Spalart, Direct simulation of a turbulent boundary layer up to Rθ = 1410, J. Fluid Mech. 187 (1988) 61–98

work page 1988

[7] [7]

T. S. Lund, X. Wu, K. D. Squires, Generation of turbulent inﬂow data for spatially-developing boundary layer simulations, J. Comput. Phys. 140 (1998) 233–258

work page 1998

[8] [8]

H. Le, P. Moin, J. Kim, Direct numerical simulation of turbulent ﬂow over a backward-facing step, J. Fluid Mech. 330 (1997) 349–374

work page 1997