Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

The balance between compactness and forecast accuracy of data-driven latent-space reduced-order models in controlled wake flows

T0 review · 3 major / 6 minor · reviewed 2026-07-31 · grok-4.5

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

Pith's one-line read For controlled wake flows, smoother POD latent dynamics beat more compact autoencoder latents on long-horizon forecasts.

desk verdict Solid empirical bake-off: for LSTM plant models on two actuated 2D wakes, POD latents beat more compact AE latents on long-horizon reliability — useful design guidance, with one untested confound on predictor capacity. read the letter →

arxiv 2607.24569 v1 pith:T6QM3DSH submitted 2026-07-27 physics.flu-dyn cs.LG

Alberto Solera-Rico , Patricia García-Caspueñas , Carlos Sanmiguel Vila , Stefano Discetti This is my paper · ORCID
classification physics.flu-dyncs.LG PACS 47.85.L47.32.Ff07.05.Mh
keywords reduced-ordermodelsactiveflowcontrolproperorthogonaldecompositionconvolutionalautoencoderslatent-spacepredictionLSTMwakeflowsmodelpredictive
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

Model-based flow control needs reduced-order models that both compress velocity fields and forecast how those reduced states evolve under actuation. This paper compares linear POD compression with nonlinear convolutional and variational autoencoders on two actuated 2D wakes (a simplified truck and the fluidic pinball), then trains LSTM predictors on each latent space. Autoencoders pack more energy into fewer coordinates and look sharper at short range, but their latent trajectories are irregular and broadband, so long-horizon forecasts degrade faster and diverge more often. POD produces smoother, nearly oscillatory latent signals that LSTMs learn and extrapolate more reliably. The practical claim is that for forecast-driven control, latent predictability can matter more than maximal compression.

What carries the argument

A two-stage latent ROM: a spatial encoder (POD versus CAE, β-VAE, or DKL-VAE) maps velocity snapshots to low-dimensional coordinates, then an LSTM-based multi-time-delay predictor steps those coordinates forward under known control inputs; end-to-end accuracy is scored on decoded fields over prediction horizon τ.

What would settle it

Retrain the same wakes with a higher-capacity real-time-feasible sequence model on identical CAE versus POD latents: if long-horizon median field accuracy and divergence rates equalize or reverse in favor of the autoencoder without violating latency or power budgets, the claim that latent stability outweighs compactness for these systems fails.

Watch

Extended reading notes

Core claim

On two controlled 2D wake datasets, nonlinear autoencoders achieve higher compression and sharper short-term reconstructions than POD, yet POD-based LSTM models deliver more accurate and stable long-horizon forecasts because POD latent trajectories are smoother and narrower-band, while autoencoder latents are irregular and broadband and therefore harder to extrapolate without catastrophic divergence.

Load-bearing premise

The paper treats the forecast gap as mainly due to the encoder by holding the temporal model family fixed to LSTMs chosen for real-time hardware limits; if a stronger sequence model closed that gap for autoencoder latents at deployable cost, the design rule would weaken.

Editorial extensions

If this is right

  • For model predictive control and forecast-based reinforcement learning on shedding wakes, prefer a slightly higher-dimensional but dynamically simpler latent basis over maximal compression.
  • Direct sequence and derivative-based LSTM predictors are better long-horizon choices than single-step autoregression for these latent ROMs.
  • Latent regularisation alone (β-VAE or decomposed-KL VAE) does not restore POD-level long-horizon reliability when the encoder is still driven by reconstruction.
  • Encoder selection for control-oriented ROMs should be scored on multi-horizon forecast stability under actuation, not only on reconstruction energy.
  • Hardware-feasible ROM design should co-select encoder and predictor rather than maximising compression first.

Reading between the lines

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

  • At higher Reynolds numbers with truly broadband turbulent wakes, the ranking may flip once linear POD needs too many modes to capture the energetically relevant content.
  • Control-conditioned or regime-local encoders could keep nonlinear compression while forcing latent trajectories onto narrower-band manifolds suited to receding-horizon optimisation.
  • Closed-loop MPC experiments that swap only the encoder under fixed compute budgets would quantify whether the longer stable horizon actually improves drag or base-pressure recovery.
  • Phase-error growth, not amplitude error, is the dominant failure mode highlighted here, so phase-aware losses or spectral penalties on the latent predictor are a natural next lever.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The manuscript studies how the choice of spatial encoder affects the long-horizon predictability of latent-space reduced-order models for actuated wake flows. Two DNS datasets (a simplified 2D truck wake at Re=500 with single-input synthetic-jet actuation, and the fluidic pinball at Re=150 with three-input cylinder rotation) are compressed with POD, a convolutional autoencoder (CAE), a β-VAE, and a decomposed-KL VAE (DKL-VAE), and the latent dynamics are forecast with three LSTM-based predictors (single-step, direct sequence, derivative-based). The central empirical finding is a compactness–forecast trade-off: the nonlinear encoders achieve higher compression (e.g., d=10 vs d=22 for the pinball at matched reconstruction energy) and sharper short-term reconstructions, but their latent trajectories are irregular and broadband, so LSTM forecasts on them degrade faster and diverge more often than on POD latents. The claim is supported by end-to-end energy metrics E(τ) over 20-seed ensembles, violin distributions exposing catastrophic-failure tails for CAE-based models, phase/amplitude error decomposition, Poincaré sections, Rosenstein-style short-horizon divergence rates, and appendices sweeping sampling rate, latent dimension, encoder architecture, and a SINDy alternative predictor.

Significance. If the result holds, it provides practically useful, falsifiable design guidance for forecast-based flow control (MPC, model-based RL): maximal compression can be counterproductive when the downstream temporal model must extrapolate the latent dynamics. The work is a careful empirical bake-off rather than a derivation, and it ships several features that deserve explicit credit: 20-seed retraining ensembles per encoder–predictor pair (so the comparison is robust to training stochasticity), an honest clipping of the reported truck horizon at τ=50 once run-to-run variability dominates, a controlled variational-regularisation test (β-VAE and DKL-VAE) showing the trade-off is not an artefact of one penalty form, and sensitivity appendices (sampling rate, latent dimension, filter count, activation) that preserve the encoder ranking. The phase/amplitude decomposition showing phase-drift-dominated long-horizon error is a nice diagnostic. The scope is limited to two 2D low-Re DNS cases with LSTM predictors, which the authors acknowledge; within that scope the evidence is solid.

major comments (3)
  1. [§2.3, §3.3, Fig. 14, §4] The headline causal attribution — that the forecast gap arises from encoder-induced irregular, broadband latent dynamics — is confounded by temporal-predictor capacity, which is never swept. §2.3 fixes the predictor family to LSTMs for a controlled comparison, and Appendix A sweeps sampling rate (A.1), latent dimension (A.2), and encoder architecture (A.3), but LSTM hidden size/layers appear in no sweep. Two internal observations make this load-bearing rather than pedantic: (i) Fig. 14 shows the encoded-reference divergence rates λref for the pinball are comparable across all four encoders (≈0.042–0.057), so the CAE latent flow is not intrinsically more sensitive to perturbations — the failure is a learnability/capacity interaction between the LSTM and the broadband latent signal, which is exactly what the 'latent-space stability' narrative (§3.3) attributes to the encoder alone; (ii) §4
  2. [§2.3, Table 3] The caption of Table 3 states that LSTM hyperparameters 'have been tuned via a manual hyperparameter search on the test set metrics.' The same test set is then used for the headline E(τ) curves and violin distributions (Figs. 7–8) that carry the paper's central conclusion. Tuning on the test set optimistically biases all reported accuracies and, more importantly for a comparative study, makes the fairness of the encoder comparison depend on equal tuning effort per architecture rather than on a held-out protocol. The 20-seed ensembles control training stochasticity but not this selection bias. This should be fixed by tuning on a validation split carved from the training data and reporting test metrics untouched; at minimum the authors must state how many configurations were tried per architecture and demonstrate that the encoder ranking is insensitive to the tuning protocol.
  3. [§3.1–§3.2, Appendix A.2, Figs. 7–8, 17] The headline comparison for the pinball contrasts POD at d=22 with CAE/β-VAE/DKL-VAE at d=10, so 'compactness versus forecast accuracy' conflates two variables: encoder class and latent dimension. Appendix A.2 sweeps d (POD d=13/22/36; CAE d=7/10/15) and reports the ranking is preserved, which addresses much of the concern, but the main-text narrative (§3.1–§3.2) should integrate this result rather than deferring it, since a reader of Figs. 7–8 alone cannot tell whether the POD advantage is a d-effect. Relatedly, a POD model truncated to d=10 (matched compactness, lower energy) would be a natural and cheap addition to Fig. 17: if d=10 POD still out-forecasts d=10 CAE, the case for latent-dynamics regularity over compression is substantially strengthened.
minor comments (6)
  1. [§2.3, Eq. (2.13)] Eq. (2.13): the control-sequence indices are inconsistent — the left argument lists [b_{t−L},…] while the latent lookback has L states [z_{t−L+1},…,z_t]; please harmonise the lookback indexing.
  2. [Fig. 6] Fig. 6 'Predicted' columns show a single representative trajectory; please state which seed/window is shown and confirm it is median-typical rather than best-case, given the heavy CAE failure tails in Figs. 7–8.
  3. [General] Typos/style: 'Corresponding author:' missing space after colon (p. 2); 'an multi-layer perceptron' → 'a multi-layer perceptron' (Fig. 5 caption); several citations have collapsed spacing ('McNallyet al.2015', 'Ceruttiet al.2020', 'Solera-Ricoet al.2024', etc.) throughout.
  4. [Fig. 14] §3.3, Fig. 14: the truck POD panel is acknowledged to lack a log-linear region, yet λref=0.075 is annotated next to values ~0.04 for the other encoders; consider omitting the annotated slope for that panel or greying it out to avoid inviting the quantitative comparison the text warns against.
  5. [§2.2.4, Table 2] The DKL-VAE regularisation weights (λ_TC differing by 10× between truck and pinball, Table 2) are stated without a tuning protocol; a sentence on how they were selected would help reproducibility, especially since DKL-VAE is presented as a controlled test of selective regularisation.
  6. [Data availability] Data availability states datasets and code 'will be made openly available… upon publication'; given the reproducibility emphasis (20-seed protocol), depositing at least the trained-model configs and analysis scripts at revision would strengthen the submission.

Circularity Check

1 steps flagged · score 1.0 of 10

Empirical encoder–predictor bake-off on held-out wakes; no derivation that redefines its target as a fit or self-citation chain.

  1. fitted input called prediction [§2.3, Table 3 note]
    "We note that the hyperparameters have been tuned via a manual hyperparameter search on the test set metrics, specifically for each predictor architecture, in order to compare each architecture in its optimal configuration."

    Predictor hyperparameters are selected using the same test metrics later used to rank encoders and architectures. That couples selection to the reported comparison and can mildly inflate absolute scores; it does not, however, make the POD-vs-CAE forecast gap true by construction, because the gap is still measured on held-out windows rather than algebraically identical to a fitted parameter.

full rationale

The paper’s load-bearing claim is comparative and empirical: on two actuated 2D wakes, CAE/β-VAE/DKL-VAE compress better and look sharper at short horizon, while POD latents yield more stable long-horizon LSTM forecasts (Figs. 7–8, §3.2–§4). Reconstruction energy E (Eq. 2.6) and end-to-end forecast E(τ) are standard L2-style metrics evaluated on held-out trajectory segments (final 10%/20% of each DNS series), not quantities defined from the same fit they are said to predict. Encoders and LSTMs are trained by ordinary reconstruction/prediction losses; the reported crossover is a measured outcome across seeds, not a forced identity. Self-citations (Solera-Rico et al. 2024/2025) supply method lineage and one dataset, but the trade-off result is produced by new side-by-side runs, not by importing a uniqueness theorem or ansatz that forbids alternatives. The only mild circularity-adjacent practice is manual hyperparameter search on test-set metrics (§2.3/Table 3), which slightly couples model selection to the evaluation split; that is leakage, not a self-definitional reduction of the scientific claim. Predictor-capacity caveats (§4) affect causal attribution strength, not circularity of the derivation. Score 1: essentially non-circular empirical work with one minor test-tuned selection step.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The paper is empirical. Load-bearing content is modeling choices and evaluation protocol rather than new physical entities. The central claim rests on DNS surrogates of controlled wakes, a finite-memory predictor hypothesis, fixed LSTM capacity as the fairness constraint, energy-threshold latent sizes, and open-loop forecast metrics as proxies for control usefulness.

free parameters (5)
  • Latent dimension d (energy targets) = truck d=8; pinball AE d=10, POD d=22
    d chosen so mean reconstruction energy is ≥95% (truck) or ≥90% (pinball); yields d=8 POD/AE truck and d=10 AE vs d=22 POD pinball.
  • β-VAE β and DKL-VAE λ_MI, λ_TC, λ_Dim = truck β=0.0025; pinball β=0.005; λ sets in Table 2
    Regularization weights hand-tuned for independence without large reconstruction loss (Table 2).
  • LSTM architecture hyperparameters (L, H, layers, widths, dropout) = see Table 3
    Per-architecture manual search on test-set metrics (explicitly stated under Table 3), so reported rankings partly reflect tuned capacity per encoder family.
  • Control excitation design (filter bands, ramp/hold schedules) = truck bandpass 0.05–0.5 t_c^{-1}; pinball discrete speeds in {-1,-0.5,0,0.5,1} with cylinder-specific ramps/holds
    Open-loop signals define the visited actuation manifold and thus which latent dynamics are observed.
  • Rosenstein-style fit windows for λ_eff = [3,10] and [0.2,10] t_c
    Divergence rates use chosen convective-time windows ([3,10] t_c truck; [0.2,10] t_c pinball), affecting quantitative slopes though mainly used ordinally.
assumptions (5)
  • domain assumption Finite-memory Mori–Zwanzig-style closure: unresolved-scale effects on latent coordinates can be approximated by an LSTM over a finite lookback L plus known controls.
    Stated as working hypothesis in §2.3; underpins treating encoder-induced latent complexity as the driver of forecast error.
  • domain assumption At Re=500 the simplified truck wake is adequately two-dimensional for this ROM study (checked with a prior 3D simulation, not shown as full analysis).
    §2.1.1; if 3D effects dominate latent predictability, the encoder ranking may change.
  • ad hoc to paper Holding the predictor family to LSTMs isolates encoder effects and matches hardware-feasible real-time constraints.
    §2.3 justification; authors later note more expressive predictors could favor AE latents (§4).
  • domain assumption Open-loop long-horizon forecast accuracy on DNS is a relevant proxy for usefulness in MPC/RL plant models.
    Framing in §1 and §4 without closed-loop experiments in this paper.
  • standard math POD optimality in L2 energy and standard AE reconstruction losses are appropriate compression objectives to compare.
    Classical snapshot POD and L2 AE training as in §2.2.
invented entities (2)
  • Operational 'latent-space stability' for controlled ROMs
    purpose: Define finite-horizon predictive reliability (bounded error, slow accumulation, preserved phase/amplitude/geometry) without claiming Lyapunov stability of the full flow.
    Introduced in §3.3 as an operational definition tailored to predictive control ROMs; diagnostic, not a new physical field.
  • DKL-VAE application as selective latent regularizer for controlled wakes independent evidence
    purpose: Test whether decomposed KL terms restore long-horizon predictability lost by CAEs.
    Builds on Chen et al. and Wang et al.; not a new physical entity, but a paper-specific regularized encoder variant in the comparison set.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The balance between compactness and forecast accuracy of data-driven latent-space reduced-order models in controlled wake flows." pith.science (2026). https://pith.science/paper/T6QM3DSH

@misc{pith2026260724569,
  author       = {Pith},
  title        = {Pith review of: The balance between compactness and forecast accuracy of data-driven latent-space reduced-order models in controlled wake flows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T6QM3DSH}},
  note         = {Machine review of arXiv:2607.24569}
}
read the original abstract

Model-based active flow control requires predictive models that are accurate, stable, and fast enough for real-time optimisation. In controlled wake flows, this is often achieved through Reduced-Order Models (ROMs) that first compress high-dimensional velocity snapshots into a latent space and then learn a time- stepping predictor for the dynamics in the latent space. Here, we study how the choice of the spatial encoder affects the predictability of the resulting latent coordinates for wake flows under control inputs. Using two actuated 2D wake configurations, a simplified truck wake and the fluidic pinball, we compare Proper Orthogonal Decomposition (POD) against nonlinear Convolutional Autoencoders (CAEs) and two types of variational autoencoders for compression, and evaluate several temporal predictors based on Long Short-Term Memory networks. CAEs achieve higher compression efficiency and sharper short-term reconstructions, but they produce latent dynamics that are more irregular and with broadband spectral content. As a consequence, long-horizon forecasts degrade faster and show a higher probability of catastrophic divergence than POD-based models. POD yields smoother latent trajectories that are easier to learn and extrapolate, leading to more reliable predictions beyond the short- term regime. These results reveal a clear trade-off between compactness and forecast accuracy, and suggest that the stability of the latent dynamics prediction can outweigh maximal compression. This is particularly relevant for control strategies rooted in forecasts of the dynamics, such as model predictive control and reinforcement learning. The findings provide practical guidance for designing actuation-aware, hardware-feasible predictive ROMs for real-time flow control.

Figures

Figures reproduced from arXiv: 2607.24569 by the authors.

Figure 1
Figure 1. Schematic of the data-driven reduced-order modelling framework. High-dimensional input velocity fields are compressed into a low-dimensional latent-space time-series using an encoder (POD or CAE). This latent state, along with the corresponding control sequence, is fed into a temporal predictor to infer the future evolution of the latent space. Finally, a decoder reconstructs the predicted latent states back into th… view at source ↗
Figure 2
Figure 2. Schematic of the simplified truck flow configuration. Zero-net-mass flow jets are illustrated in blue. The control is implemented using two opposing flow jets at the sides of the vehicle base with zero-net mass flow, similar to previous experimental setups (Barros et al. 2016; Cerutti et al. 2020). In this case, the overall mass flow rate of the two jets combined is forced to be zero at each instant of the simulatio… view at source ↗
Figure 3
Figure 3. illustrates the global mesh and a flow sample around the jet region [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: The geometry consists of three cylinders of equal diameter 𝐷 = 2𝑅, with their centres located at the vertices of an equilateral triangle. The centre-to-centre side length of the triangle is 3𝑅. The origin of the reference system is located in the midpoint between the t…
Figure 5
Figure 5. Figure 5: Schematic of the three LSTM-based prediction architectures tested. For all architectures, the latent space (in blue) and control space (in gray) are fed as input throughout the entire lookback period up to the current time step 𝑡. (Top) Single-step model , consisting o…
Figure 6
Figure 6. Figure 6: Qualitative comparison of flow field reconstruction and long-term prediction for the fluidic pinball (left, at 𝜏 = 20) and the simplified truck (right, at 𝜏 = 50). The “True” row shows the reference nondimensional vorticity; all rows for a given case are extracted from…
Figure 7
Figure 7. Figure 7: Prediction accuracy 𝐸 for the fluidic pinball over an ensemble of 20 training seeds. Left: median of 𝐸 (Eq. 2.6) as a function of the prediction horizon 𝜏 (convective time 𝑡𝑐 units), taken over all prediction windows in the test set and all seeds. Colours denote the LS…
Figure 8
Figure 8. Figure 8: Prediction accuracy 𝐸 for the simplified truck over an ensemble of 20 training seeds, with the same protocol, statistics and line conventions as [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]
Figure 9
Figure 9. Figure 9: Temporal evolution of the first four leading latent coefficients (energy-ordered for POD; ordered by peak Power Spectral Density (PSD) of the encoded reference for CAE, 𝛽-VAE and DKL-VAE), comparing the ground truth (solid black line) with the prediction of the best-pe…
Figure 10
Figure 10. Figure 10: Pointwise normalised 𝐿2 error magnitude of the predicted velocity fluctuations, combining the streamwise and crosswise components as √︁ 𝑒 2 𝑢 + 𝑒 2 𝑣 at each grid point, where each component 𝑒𝑖 = ⟨(𝑢ˆ𝑖 − 𝑢𝑖) 2 ⟩/𝜎 2 𝑖 is the mean squared error normalised by the varian…
Figure 11
Figure 11. Figure 11: Phase portraits of the dominant standardised latent pair (𝑧 ∗ 1 , 𝑧∗ 2 ) for the encoded reference (black) and the Direct sequence LSTM prediction (coloured). Rows: truck (top) and fluidic pinball (bottom). Columns: POD, 𝛽-VAE , CAE and DKL-VAE. Standardisation uses t…
Figure 12
Figure 12. Figure 12: Poincare sections in ´ (𝑧 ∗ 2 , 𝑧∗ 3 ) taken at positive crossings of 𝑧 ∗ 1 = 0. Encoded reference (black) versus LSTM-predicted (coloured), with the same row/column layout as [PITH_FULL_IMAGE:figures/full_fig_p021_12.png]
Figure 13
Figure 13. Figure 13: Absolute phase error |𝛥𝜙| (left) and absolute amplitude error |𝛥𝐴| (right) of the dominant latent pair, for the truck (top) and fluidic pinball (bottom). Lines: POD (blue), 𝛽-VAE (green), CAE (orange) and DKL-VAE (pink). Thick lines are a centred moving average over o…
Figure 14
Figure 14. Figure 14: Rosenstein-style logarithmic separation of initially close latent-state pairs, ⟨ln 𝑑(𝑡)⟩, for the encoded reference (solid) and the LSTM-predicted trajectory (dashed). Rows: truck and fluidic pinball cases. Columns: POD, 𝛽-VAE , CAE and DKL-VAE. The dotted black line …
Figure 15
Figure 15. Figure 15: Effect of the sampling rate of the latent-space time series on the prediction performance for the fluidic pinball with 𝐸 = 90% and the Direct sequence LSTM predictor. Three rates are compared: 𝛥𝑡, 2𝛥𝑡 and 4𝛥𝑡. (Left) Prediction accuracy 𝐸 as a function of the predicti…
Figure 16
Figure 16. Figure 16: Power spectral density against Strouhal number St = 𝑓 𝐿𝑟 𝑒 𝑓 /𝑈∞ of the four leading latent coordinates (𝑧1–𝑧4) for the encoded reference (solid) and the Direct sequence LSTM prediction (dashed), with 𝐿𝑟 𝑒 𝑓 the truck width or cylinder diameter and 𝑈∞ the reference ve…
Figure 17
Figure 17. Figure 17: Increasing the reconstructed energy reduces the prediction variance for short-to-medium horizons (up to approximately 𝜏 = 5). At long horizons, however, the baseline 𝐸 = 90% remains the most reliable configuration for both encoders. Larger latent dimensions improve sh…
Figure 18
Figure 18. Figure 18: Median prediction accuracy 𝐸(𝜏) over the training-seed ensemble (left) and pooled distributions of the instantaneous 𝐸 at 𝜏 = 0.1 and 𝜏 = 5 (right) for the paper baselines (POD22, CAE1, 𝛽-VAE, DKL-VAE) and the architecture variants, evaluated on the fluidic pinball ca…
Figure 19
Figure 19. Figure 19: Comparison between SINDy and the Direct sequence LSTM predictor on the fluidic pinball with 𝐸 = 90%. (Left) Prediction accuracy 𝐸 as a function of the prediction horizon 𝜏. (Right) Violin plots of 𝐸 at a short-term and a long-term horizon. White markers indicate the m…
Figure 20
Figure 20. Figure 20: Comparison between SINDy and the Direct sequence LSTM predictor on the truck case. (Left) Prediction accuracy 𝐸 as a function of the prediction horizon 𝜏. (Right) Violin plots of 𝐸 at a short-term and a long-term horizon. White markers indicate the median (◦ for POD, …

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Autoregressive rollout error in latent-space reduced-order models of bluff-body wakes is accumulated phase drift

    physics.flu-dyn 2026-08 conditional novelty 6.0 of 10

    Autoregressive rollout error in latent-space wake models is dominated by a linear drift in the oscillation phase, and can be corrected offline with one fitted parameter per latent coordinate.

Reference graph

Works this paper leans on

102 extracted references · 5 canonical work pages · cited by 1 Pith paper

  1. [1]

    International Journal of Heat and Fluid Flow , volume=

    Deep-reinforcement-learning-based separation control in a two-dimensional airfoil , author=. International Journal of Heat and Fluid Flow , volume=. 2025 , publisher=

    2025

  2. [2]

    Communications Engineering , volume=

    Flow control of three-dimensional cylinders transitioning to turbulence via multi-agent reinforcement learning , author=. Communications Engineering , volume=. 2025 , publisher=

    2025

  3. [3]

    arXiv preprint arXiv:2510.11600 , year=

    A framework for realisable data-driven active flow control using model predictive control applied to a simplified truck wake , author=. arXiv preprint arXiv:2510.11600 , year=

    arXiv

  4. [4]

    AIAA Journal , pages=

    Model-based reinforcement learning for control of strongly disturbed unsteady aerodynamic flows , author=. AIAA Journal , pages=. 2025 , publisher=

    2025

  5. [5]

    Journal of Fluid Mechanics , volume=

    Machine-learned control-oriented flow estimation for multi-actuator multi-sensor systems exemplified for the fluidic pinball , author=. Journal of Fluid Mechanics , volume=. 2022 , publisher=

    2022

  6. [6]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  7. [7]

    arXiv preprint arXiv:2502.14819 , year=

    Learning from Reward-Free Offline Data: A Case for Planning with Latent Dynamics Models , author=. arXiv preprint arXiv:2502.14819 , year=

    arXiv

  8. [8]

    2023 , issn =

    Gym-preCICE: Reinforcement learning environments for active flow control , journal =. 2023 , issn =. doi:10.1016/j.softx.2023.101446 , author =

    arXiv 2023

Show all 102 references
  1. [9]

    OpenFOAM® Journal , volume =

    Chourdakis, Gerasimos and Schneider, David and Uekermann, Benjamin , title =. OpenFOAM® Journal , volume =. 2023 , month =

    2023

  2. [10]

    Open Research Europe , volume =

    Chourdakis, G and Davis, K and Rodenberg, B and Schulte, M and Simonis, F and Uekermann, B and Abrams, G and Bungartz, HJ and Cheung Yau, L and Desai, I and Eder, K and Hertrich, R and Lindner, F and Rusch, A and Sashko, D and Schneider, D and Totounferoush, A and Volland, D a...

    2022

  3. [11]

    and Ross, James C

    Storms, Bruce L. and Ross, James C. and Heineck, James T. and Walker, Stephen M. and Driver, David M. and Zilliac, Gregory G. and Bencze, Daniel P. , title =. 2001 , number =

    2001

  4. [12]

    Advances in neural information processing systems , volume=

    A unified approach to interpreting model predictions , author=. Advances in neural information processing systems , volume=

  5. [13]

    and Cornejo Maceda, G

    Castellanos, R. and Cornejo Maceda, G. Y. and de la Fuente, I. and Noack, B. R. and Ianiro, A. and Discetti, S. , title =. Physics of Fluids , volume =. 2022 , month =

    2022

  6. [14]

    Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control , volume=

    Rabault, Jean and Kuchta, Miroslav and Jensen, Atle and Réglade, Ulysse and Cerardi, Nicolas , year=. Artificial neural networks trained through deep reinforcement learning discover control strategies for active flow control , volume=. doi:10.1017/jfm.2019.62 , journal=

    2019 doi

  7. [15]

    Physical Review Fluids , volume =

    Neural networks in feedback for flow analysis and control , author =. Physical Review Fluids , volume =. 2024 , month =

    2024

  8. [16]

    Self-tuning model predictive control for wake flows , volume =

    Marra, Luigi and Meil. Self-tuning model predictive control for wake flows , volume =. Journal of Fluid Mechanics , issn =. 2024 , publisher =

    2024

  9. [17]

    Journal of Fluid Mechanics , volume=

    Actuation manifold from snapshot data , author=. Journal of Fluid Mechanics , volume=. 2024 , publisher=

    2024

  10. [18]

    Theoretical and computational fluid dynamics , volume=

    Deep model predictive flow control with limited sensor data and online learning , author=. Theoretical and computational fluid dynamics , volume=. 2020 , publisher=

    2020

  11. [19]

    2022 , eprint=

    VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning , author=. 2022 , eprint=

    2022

  12. [20]

    Applied Mechanics Reviews , volume=

    Closed-loop turbulence control: Progress and challenges , author=. Applied Mechanics Reviews , volume=. 2015 , publisher=

    2015

  13. [21]

    Physics of Fluids , volume=

    Assessment of unsteady flow predictions using hybrid deep learning based reduced-order models , author=. Physics of Fluids , volume=. 2021 , publisher=

    2021

  14. [22]

    Long Short-Term Memory , year=

    Hochreiter, Sepp and Schmidhuber, Jürgen , journal=. Long Short-Term Memory , year=

  15. [23]

    Advances in neural information processing systems , volume=

    Pytorch: An imperative style, high-performance deep learning library , author=. Advances in neural information processing systems , volume=

  16. [24]

    Fluids , VOLUME =

    Kim, Innyoung and Jeon, Youngmin and Chae, Jonghyun and You, Donghyun , TITLE =. Fluids , VOLUME =. 2024 , NUMBER =

    2024

  17. [25]

    2021 , issn =

    A review on deep reinforcement learning for fluid mechanics , journal =. 2021 , issn =. doi:10.1016/j.compfluid.2021.104973 , author =

    2021

  18. [26]

    Triantafyllou and George Em Karniadakis , title =

    Dixia Fan and Liu Yang and Zhicheng Wang and Michael S. Triantafyllou and George Em Karniadakis , title =. Proceedings of the National Academy of Sciences , volume =. 2020 , doi =

    2020

  19. [27]

    Nature Communications , volume =

    Deep reinforcement learning for active flow control in a turbulent separation bubble , author =. Nature Communications , volume =. 2025 , doi =

    2025

  20. [28]

    Physics of Fluids , volume=

    Reduced-order modeling of advection-dominated systems with recurrent neural networks and convolutional autoencoders , author=. Physics of Fluids , volume=. 2021 , publisher=

    2021

  21. [29]

    Closed-loop supersonic flow control with a high-speed experimental deep reinforcement learning framework , volume=

    Zong, Haohua and Wu, Yun and Li, Jinping and Su, Zhi and Liang, Hua , year=. Closed-loop supersonic flow control with a high-speed experimental deep reinforcement learning framework , volume=. doi:10.1017/jfm.2025.160 , journal=

    2025 doi

  22. [30]

    European Physical Journal E , volume =

    Deep reinforcement learning for turbulent drag reduction in channel flows , author =. European Physical Journal E , volume =. 2023 , doi =

    2023

  23. [31]

    and Rabault, J

    Vignon, C. and Rabault, J. and Vinuesa, R. , title =. Physics of Fluids , volume =. 2023 , month =

    2023

  24. [32]

    Advances in neural information processing systems , volume=

    Differentiable mpc for end-to-end planning and control , author=. Advances in neural information processing systems , volume=

  25. [33]

    2017 , publisher =

    Model Predictive Control: Theory, Computation, and Design , author =. 2017 , publisher =

    2017

  26. [34]

    and Sheplak, Mark

    Cattafesta, Louis N. and Sheplak, Mark. Actuators for Active Flow Control. Annual Review of Fluid Mechanics. 2011. doi:10.1146/annurev-fluid-122109-160634

    2011 doi

  27. [35]

    SYNTHETIC JETS

    Glezer, Ari and Amitay, Michael. SYNTHETIC JETS. Annual Review of Fluid Mechanics. 2002. doi:10.1146/annurev.fluid.34.090501.094913

    2002

  28. [36]

    2000 , issn =

    The control of flow separation by periodic excitation , journal =. 2000 , issn =. doi:10.1016/S0376-0421(00)00008-7 , author =

    2000 doi

  29. [37]

    and Noack, Bernd R

    Brunton, Steven L. and Noack, Bernd R. and Koumoutsakos, Petros. Machine Learning for Fluid Mechanics. Annual Review of Fluid Mechanics. 2020. doi:10.1146/annurev-fluid-010719-060214

    2020 doi

  30. [38]

    Experiments in fluids , volume=

    Aerodynamic drag reduction of a simplified squareback vehicle using steady blowing , author=. Experiments in fluids , volume=. 2012 , publisher=

    2012

  31. [39]

    Journal of Wind Engineering and Industrial Aerodynamics , volume=

    Drag reduction on a flat-back ground vehicle with active flow control , author=. Journal of Wind Engineering and Industrial Aerodynamics , volume=. 2015 , publisher=

    2015

  32. [40]

    Fluids , volume=

    Active flow control on a square-back road vehicle , author=. Fluids , volume=. 2020 , publisher=

    2020

  33. [41]

    Physics of Fluids , volume=

    Deep reinforcement learning for active control of a three-dimensional bluff body wake , author=. Physics of Fluids , volume=. 2022 , publisher=

    2022

  34. [42]

    Physics of Fluids , volume=

    Flow topology of deep reinforcement learning drag-reduced bluff body wakes , author=. Physics of Fluids , volume=. 2024 , publisher=

    2024

  35. [43]

    Journal of Fluid Mechanics , volume=

    Bluff body drag manipulation using pulsed jets and Coanda effect , author=. Journal of Fluid Mechanics , volume=. 2016 , publisher=

    2016

  36. [44]

    Nature Communications , volume=

    Learning the intrinsic dynamics of spatio-temporal processes through Latent Dynamics Networks , author=. Nature Communications , volume=. 2024 , publisher=

    2024

  37. [45]

    Theoretical and Computational Fluid Dynamics , volume=

    Reducing the pressure drag of a D-shaped bluff body using linear feedback control , author=. Theoretical and Computational Fluid Dynamics , volume=. 2017 , publisher=

    2017

  38. [46]

    Journal of fluid mechanics , volume=

    Feedback shear layer control for bluff body drag reduction , author=. Journal of fluid mechanics , volume=. 2008 , publisher=

    2008

  39. [47]

    Proceedings of the national academy of sciences , volume=

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems , author=. Proceedings of the national academy of sciences , volume=. 2016 , publisher=

    2016

  40. [48]

    Proceedings of the Royal Society A , volume=

    Sparse identification of nonlinear dynamics for model predictive control in the low-data limit , author=. Proceedings of the Royal Society A , volume=. 2018 , publisher=

    2018

  41. [49]

    Nature Communications , volume=

    -variational autoencoders and transformers for reduced-order modelling of fluid flows , author=. Nature Communications , volume=. 2024 , publisher=

    2024

  42. [50]

    Advances in Neural Information Processing Systems , volume=

    Isolating sources of disentanglement in variational autoencoders , author=. Advances in Neural Information Processing Systems , volume=

  43. [51]

    2026 , issn =

    Information decomposition for disentangled and interpretable manifold learning of fluid flows via variational autoencoders , journal =. 2026 , issn =. doi:10.1016/j.cja.2026.104342 , author =

    2026

  44. [52]

    Journal of Fluid Mechanics , volume=

    Robust flow control and optimal sensor placement using deep reinforcement learning , author=. Journal of Fluid Mechanics , volume=. 2021 , publisher=

    2021

  45. [53]

    L. S. Shapley , publisher =. A Value for n-Person Games , booktitle =. doi:10.1515/9781400881970-018 , isbn =

    doi

  46. [54]

    International Journal of Heat and Fluid Flow , volume=

    Additive-feature-attribution methods: A review on explainable artificial intelligence for fluid dynamics and heat transfer , author=. International Journal of Heat and Fluid Flow , volume=. 2025 , publisher=

    2025

  47. [55]

    Nature Communications , volume=

    Grasping extreme aerodynamics on a low-dimensional manifold , author=. Nature Communications , volume=. 2023 , publisher=

    2023

  48. [56]

    doi:10.1088/1873-7005/ade8a2 , year =

    Fukagata, Koji and Fukami, Kai , title =. doi:10.1088/1873-7005/ade8a2 , year =

    doi

  49. [57]

    AIAA Journal , year=

    Model-based reinforcement learning for control of strongly disturbed unsteady aerodynamic flows , author=. AIAA Journal , year=

  50. [58]

    AIAA Journal , pages=

    Information-theoretic machine learning for time-varying mode decomposition of separated aerodynamic flows , author=. AIAA Journal , pages=. 2025 , publisher=

    2025

  51. [59]

    Journal of Fluid Mechanics , volume=

    Observable-augmented manifold learning for multi-source turbulent flow data , author=. Journal of Fluid Mechanics , volume=. 2025 , publisher=

    2025

  52. [60]

    arXiv preprint arXiv:2506.12777 , year=

    Phase autoencoder for rapid data-driven synchronization of rhythmic spatiotemporal patterns , author=. arXiv preprint arXiv:2506.12777 , year=

  53. [61]

    2001 , publisher=

    Bewley, Thomas R and Moin, Parviz and Temam, Roger , journal=. 2001 , publisher=

    2001

  54. [62]

    Advances in Neural Information Processing Systems , volume=

    Deep dynamical modeling and control of unsteady fluid flows , author=. Advances in Neural Information Processing Systems , volume=

  55. [63]

    International conference on machine learning , pages=

    Axiomatic attribution for deep networks , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  56. [64]

    International Journal of Control, Automation and Systems , volume=

    Model predictive control: Review of the three decades of development , author=. International Journal of Control, Automation and Systems , volume=. 2011 , publisher=

    2011

  57. [65]

    Camacho and Carlos Bordons , title =

    Eduardo F. Camacho and Carlos Bordons , title =. 2013 , isbn =. doi:10.1007/978-0-85729-398-5 , number =

    2013 doi

  58. [66]

    IEEE Access , volume =

    Blaud, Pierre Cl. IEEE Access , volume =. 2023 , publisher =

    2023

  59. [67]

    and Morzyński, Marek and Pastur, Luc R

    Deng, Nan and Noack, Bernd R. and Morzyński, Marek and Pastur, Luc R. , year=. Low-order model for successive bifurcations of the fluidic pinball , volume=. doi:10.1017/jfm.2019.959 , journal=

    2019 doi

  60. [68]

    2025 , issn =

    Nonlinear dynamics of the actuated fluidic pinball — Steady, periodic, and chaotic regimes , journal =. 2025 , issn =. doi:10.1016/j.chaos.2025.116075 , author =

    2025

  61. [69]

    arXiv preprint arXiv:1511.07289 , volume=

    Fast and accurate deep network learning by exponential linear units (elus) , author=. arXiv preprint arXiv:1511.07289 , volume=

    arXiv

  62. [70]

    G. E. Hinton and R. R. Salakhutdinov , title =. Science , volume =. 2006 , doi =

    2006

  63. [71]

    and Bottou, L

    Lecun, Y. and Bottou, L. and Bengio, Y. and Haffner, P. , journal=. Gradient-based learning applied to document recognition , year=

  64. [72]

    arXiv preprint arXiv:1412.6980 , year=

    Adam: A method for stochastic optimization , author=. arXiv preprint arXiv:1412.6980 , year=

    arXiv

  65. [73]

    Turbulence and the dynamics of coherent structures

    Sirovich, Lawrence , journal=. Turbulence and the dynamics of coherent structures

  66. [74]

    Journal of Computational Design and Engineering , volume =

    Yang, Sunwoong and Vinuesa, Ricardo and Kang, Namwoo , title =. Journal of Computational Design and Engineering , volume =. 2025 , month =

    2025

  67. [75]

    Physics of Fluids , volume =

    Robledo, Isaac and Alfaro, Juan and Duro, Víctor and Solera-Rico, Alberto and Castellanos, Rodrigo and Sanmiguel Vila, Carlos , title =. Physics of Fluids , volume =. 2026 , month =

    2026

  68. [76]

    2022 , issn =

    Towards extraction of orthogonal and parsimonious non-linear modes from turbulent flows , journal =. 2022 , issn =. doi:10.1016/j.eswa.2022.117038 , author =

    2022

  69. [77]

    2016 , issn =

    On the near wake of a simplified heavy vehicle , journal =. 2016 , issn =. doi:10.1016/j.jfluidstructs.2016.07.011 , author =

    2016 doi

  70. [78]

    Physics of Fluids , volume =

    Fukami, Kai and Nakamura, Taichi and Fukagata, Koji , title =. Physics of Fluids , volume =. 2020 , month =

    2020

  71. [79]

    Nonlinear mode decomposition with convolutional neural networks for fluid dynamics , volume=

    Murata, Takaaki and Fukami, Kai and Fukagata, Koji , year=. Nonlinear mode decomposition with convolutional neural networks for fluid dynamics , volume=. doi:10.1017/jfm.2019.822 , journal=

    2019 doi

  72. [80]

    Physics of Fluids , volume =

    Zhu, Yin and Sun, Qiangqiang and Xiao, Dandan and Yao, Jie and Mao, Xuerui , title =. Physics of Fluids , volume =. 2024 , month =

    2024

  73. [81]

    Irina Higgins and Loic Matthey and Arka Pal and Christopher Burgess and Xavier Glorot and Matthew Botvinick and Shakir Mohamed and Alexander Lerchner , booktitle=. beta-. 2017 , url=

    2017

  74. [82]

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , year =

    Gupta, Priyam and Schmid, Peter and Sipp, Denis and Sayadi, Taraneh and Rigas, Georgios , title =. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences , year =

  75. [83]

    Nathan and Brunton, Steven L

    Lusch, Bethany and Kutz, J. Nathan and Brunton, Steven L. , title =. Nature Communications , year =

  76. [84]

    Computer Methods in Applied Mechanics and Engineering , year =

    Menier, Emmanuel and Bucci, Michele Alessandro and Yagoubi, Mouadh and Mathelin, Lionel and Schoenauer, Marc , title =. Computer Methods in Applied Mechanics and Engineering , year =

  77. [85]

    and Afanasiev, Konstantin and Morzy

    Noack, Bernd R. and Afanasiev, Konstantin and Morzy. A hierarchy of low-dimensional models for the transient and post-transient cylinder wake , journal =. 2003 , volume =

    2003

  78. [86]

    and Girvan, Michelle and Ott, Edward , title =

    Pathak, Jaideep and Lu, Zhixin and Hunt, Brian R. and Girvan, Michelle and Ott, Edward , title =. Chaos: An Interdisciplinary Journal of Nonlinear Science , year =

  79. [87]

    and Sapsis, Themistoklis P

    Vlachas, Pantelis-Rafail and Pathak, Jaideep and Hunt, Brian R. and Sapsis, Themistoklis P. and Girvan, Michelle and Ott, Edward and Koumoutsakos, Petros , title =. Neural Networks , year =

  80. [88]

    1993 , issn =

    Physica D: Nonlinear Phenomena , volume =. 1993 , issn =. doi:https://doi.org/10.1016/0167-2789(93)90009-P , author =

    1993 doi

  81. [89]

    Journal of fluid mechanics , volume=

    Spectral analysis of nonlinear flows , author=. Journal of fluid mechanics , volume=. 2009 , publisher=. doi:10.1017/S0022112009992059 , abstract =

    2009 doi

  82. [90]

    Analysis of fluid flows via spectral properties of the Koopman operator

    Mezić, Igor. Analysis of fluid flows via spectral properties of the Koopman operator. Annual Review of Fluid Mechanics. 2013. doi:https://doi.org/10.1146/annurev-fluid-011212-140652

    2013 doi

  83. [91]

    and Dawson, Scott T

    Taira, Kunihiko and Brunton, Steven L. and Dawson, Scott T. M. and Rowley, Clarence W. and Colonius, Tim and McKeon, Beverley J. and Schmidt, Oliver T. and Gordeyev, Stanislav and Theofilis, Vassilios and Ukeiley, Lawrence S. , title =. AIAA Journal , volume =. 2017 , doi =

    2017

  84. [92]

    and Budi

    Brunton, Steven L. and Budi. SIAM Review , volume =. 2022 , doi =

    2022

  85. [93]

    2018 , issn =

    Linear predictors for nonlinear dynamical systems: Koopman operator meets model predictive control , journal =. 2018 , issn =. doi:https://doi.org/10.1016/j.automatica.2018.03.046 , author =

    2018 doi

  86. [94]

    2021 , publisher=

    Fresca, Stefania and Dede’, Luca and Manzoni, Andrea , journal=. 2021 , publisher=. doi:https://doi.org/10.1007/s10915-021-01462-7 , abstract=

    2021 doi

  87. [95]

    2022 , issn =

    Computer Methods in Applied Mechanics and Engineering , volume =. 2022 , issn =. doi:https://doi.org/10.1016/j.cma.2021.114181 , author =

    2022

  88. [96]

    Journal of Fluid Mechanics , volume=

    Sparse identification of nonlinear dynamics with low-dimensionalized flow representations , author=. Journal of Fluid Mechanics , volume=. 2021 , publisher=

    2021

  89. [97]

    2016 , publisher=

    Brunton, Steven L and Proctor, Joshua L and Kutz, J Nathan , journal=. 2016 , publisher=

    2016

  90. [98]

    Acta Mathematica , number =

    Torsten Carleman , title =. Acta Mathematica , number =. 1932 , doi =

    1932

  91. [99]

    The need for a pressure-term representation in empirical

    Noack, Bernd R and Papas, Paul and Monkewitz, Peter A , journal=. The need for a pressure-term representation in empirical. 2005 , publisher=

    2005

  92. [100]

    2025 , month =

    Stability analysis of chaotic systems in latent spaces , journal =. 2025 , month =. doi:10.1007/s11071-024-10712-w , issn =

    2025 doi

  93. [101]

    Synchronization in chaotic systems , author =. Phys. Rev. Lett. , volume =. 1990 , month =

    1990

  94. [102]

    2022 , issn =

    Machine learning-based reduced-order modeling and predictive control of nonlinear processes , journal =. 2022 , issn =. doi:10.1016/j.cherd.2022.02.005 , author =

    2022 doi

Pith tools

Reviewed July 31, 2026 · model on record in the stance chip above.