pith. sign in

arxiv: 2510.25625 · v2 · submitted 2025-10-29 · ⚛️ physics.flu-dyn

MoTIF: A Mode-Structured Tensor Framework for Multi-Parametric Approximation, Super-Resolution and Forecasting of Unsteady Systems

Guillermo Barrag\'an , Ashton Hetherington , Arindam Sengupta , Rodrigo Abad\'ia-Heredia , Jes\'us Garicano-Mena , Soledad Le Clainche This is my paper

Pith reviewed 2026-05-18 03:03 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords tensor decompositionHOSVDreduced-order modelingunsteady flowsGaussian process regressionrecurrent neural networksmulti-parametric approximationflow forecasting
0
0 comments X

The pith

MoTIF separates flow data into parameter, spatial, and temporal modes to reconstruct unseen configurations and forecast evolution with errors below 2%.

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

The paper introduces a framework called MoTIF that uses high-order singular value decomposition to break down multi-dimensional unsteady flow datasets into separate modes for physical parameters, space, and time. This separation allows applying Gaussian Process Regression to handle parametric and spatial approximations for completing databases and enhancing resolution, while recurrent neural networks predict the temporal evolution. A sympathetic reader would care because it offers a non-intrusive way to model complex high-dimensional systems without needing full high-fidelity simulations for every parameter combination. The validation on laminar flows with varying Reynolds numbers and attack angles shows consistent accuracy.

Core claim

The central discovery is that a mode-structured tensor approach based on HOSVD enables decoupled learning of operators on each mode, preserving the tensor structure while achieving accurate multi-parametric approximation, super-resolution, and temporal forecasting in unsteady laminar flows, with relative root mean square errors below 2% compared to high-fidelity simulations.

What carries the argument

High-Order Singular Value Decomposition (HOSVD) to obtain a structured multilinear representation separating physical parameters, spatial coordinates, and temporal evolution into distinct modal components, followed by dedicated operators like Gaussian Process Regression and recurrent neural networks on each mode.

If this is right

  • Accurate reconstruction of flow fields for parameter values not in the original database.
  • Enhancement of spatial resolution in the approximated solutions.
  • Forecasting of future time steps in the system evolution.
  • Provision of a scalable reduced-order modeling alternative for parametric dynamical systems.

Where Pith is reading between the lines

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

  • Similar mode separation could apply to other unsteady systems in engineering, such as structural vibrations or heat transfer problems.
  • Integrating additional physical constraints into the modal operators might further reduce errors in complex regimes.
  • The success with low errors suggests the method could reduce computational costs significantly for parametric studies in fluid dynamics.

Load-bearing premise

The unsteady flow data admits a sufficiently low-rank multilinear structure that permits independent approximation operators on the separated parameter, spatial, and temporal modes without destroying essential coupling.

What would settle it

Running the MoTIF framework on a new set of unsteady flow simulations with Reynolds numbers and angles of attack outside the training set and measuring if the relative root mean square error exceeds 2% for reconstructions or predictions.

Figures

Figures reproduced from arXiv: 2510.25625 by Arindam Sengupta, Ashton Hetherington, Guillermo Barrag\'an, Jes\'us Garicano-Mena, Rodrigo Abad\'ia-Heredia, Soledad Le Clainche.

Figure 1
Figure 1. Figure 1: Sketch of the methodology applied for the SVD-based alignment of the databases obtained [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Sketch of the methodology applied for the development of the proposed [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Sketch of the methodology applied for the generation of databases unseen flow conditions. [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sketch of the methodology applied for the resolution enhancement using GPR. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Sketch of the methodology used for forecasting. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Design space distribution, each black dot represents a database obtained through numerical [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Streamwise velocity field of the database at different flow conditions: (a) Re = 200 & AoA [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Sketch of the computational domain dimensions for the two-dimensional numerical simula [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Structured mesh developed for the two-dimensional numerical simulation of the laminar [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Sketch of the computational domain dimensions and boundary conditions. A fixed [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Singular value decay (a), and cumulative energy (b) plots for the SVD mode matrix [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Same as Fig. 11a, but for the AoA SVD mode matrix [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: GPR and piecewise linear interpolation for the first 3 SVD modes calculated associated to [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: compares the results obtained by applying GPR and piecewise linear interpolation along each of the retained SVD modes associated with the AoA dimension. It can be observed that in the first SVD mode (Fig. 14a), the data distribution shows a linear trend with a negative slope, where both the regression and interpolation methods produce similar results. Similar to the first mode, the second mode shows initi… view at source ↗
Figure 15
Figure 15. Figure 15: Singular value decomposition (SVD) results for both X and Y dimensions. (a,b) show the [PITH_FULL_IMAGE:figures/full_fig_p021_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Same as Fig. 11a, but for the temporal SVD mode matrix [PITH_FULL_IMAGE:figures/full_fig_p022_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Forecast of the LSTM-based RNN on each of the 7 modes retained along the temporal [PITH_FULL_IMAGE:figures/full_fig_p023_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Design space (in black) and interpolated data (in red) for the filling missing data from [PITH_FULL_IMAGE:figures/full_fig_p024_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: FAoA task; High-resolution prediction results obtained using the proposed HybriNet, with [PITH_FULL_IMAGE:figures/full_fig_p024_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: FRe task; Same as 19 but with Re = 260 and AoA = 20◦ . The evolution of the streamwise velocity at a point of interest (x = 70, y = 80), located in the wake region behind the bluff body, is shown in [PITH_FULL_IMAGE:figures/full_fig_p025_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: FAoA task (Re = 240 and AoA = 15◦ ); Temporal evolution of the streamwise velocity component in a point of interest of the predicted database. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: FRe task; Same as 21 but with Re = 260 and AoA = 20◦ [PITH_FULL_IMAGE:figures/full_fig_p026_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Fill inconsistent databases scenario, FRe test case (Re = 240, AoA = 15 [PITH_FULL_IMAGE:figures/full_fig_p026_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: Same as 23 but for the FRe case, with Re = 260 and AoA = 20◦ . 4.5 Generation of databases for unseen flow conditions. This section presents the results obtained for generating data under unseen flow conditions using the proposed HybriNet. Two distinct cases were considered, with the objective of predicting flow conditions where neither the AoA nor the Reynolds number exist in the original database: gener… view at source ↗
Figure 25
Figure 25. Figure 25: Same as Fig [PITH_FULL_IMAGE:figures/full_fig_p027_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: N1 task; Same as 19 but with Re = 230 and AoA = 22.5 ◦ . The predictions obtained for the N2 test case are shown in [PITH_FULL_IMAGE:figures/full_fig_p028_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: N2 task; Same as 19 but with Re = 245 and AoA = 11◦ . Figures 28 and 29 show the evolution of the streamwise velocity magnitude at a point of interest 28 [PITH_FULL_IMAGE:figures/full_fig_p028_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: N1 task; Same as 21 but with Re = 230 and AoA = 22.5 ◦ [PITH_FULL_IMAGE:figures/full_fig_p029_28.png] view at source ↗
Figure 29
Figure 29. Figure 29: N2 task; Same as 19 but with Re = 245 and AoA = 11◦ [PITH_FULL_IMAGE:figures/full_fig_p029_29.png] view at source ↗
Figure 30
Figure 30. Figure 30: N1 task; Same as 23 but with Re = 230 and AoA = 22.5 ◦ . The normalized error distribution for the N2 test case is presented as relative frequency histograms for the streamwise and normal velocity components in [PITH_FULL_IMAGE:figures/full_fig_p030_30.png] view at source ↗
Figure 31
Figure 31. Figure 31: N2 task, Same as 23 but with Re = 245 and AoA = 11◦ . 4.6 Summary of the results obtained [PITH_FULL_IMAGE:figures/full_fig_p030_31.png] view at source ↗
read the original abstract

We introduce MoTIF, a mode-structured tensor framework for multi-parametric approximation, super-resolution, and temporal forecasting of high-dimensional unsteady systems. The methodology leverages High-Order Singular Value Decomposition (HOSVD) to obtain a structured multilinear representation of multi-dimensional datasets, separating physical parameters, spatial coordinates, and temporal evolution into distinct modal components. This decomposition enables the application of dedicated approximation operators to each mode. Gaussian Process Regression is employed to interpolate and extrapolate parametric and spatial modal matrices, enabling database completion and resolution enhancement, while recurrent neural networks are applied to the temporal mode to forecast system evolution. This decoupled operator-learning strategy preserves the intrinsic tensor structure while providing a flexible non-intrusive reduced-order modelling framework. The proposed methodology is validated on a database of unsteady laminar flow simulations with varying Reynolds numbers and angles of attack. Accurate reconstruction of unseen flow configurations and temporal prediction are achieved, with relative root mean square errors consistently below 2\% compared to high-fidelity simulations. The framework provides a scalable and mathematically structured alternative to conventional surrogate modelling approaches for high-dimensional parametric dynamical systems.

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 paper introduces MoTIF, a mode-structured tensor framework for multi-parametric approximation, super-resolution, and temporal forecasting of high-dimensional unsteady systems. It employs High-Order Singular Value Decomposition (HOSVD) to obtain a structured multilinear representation separating physical parameters, spatial coordinates, and temporal evolution into distinct modal components. Dedicated operators are then applied independently: Gaussian Process Regression (GPR) for interpolating and extrapolating parametric and spatial modal matrices, and recurrent neural networks (RNNs) for forecasting the temporal mode. The framework is validated on a database of unsteady laminar flow simulations with varying Reynolds numbers and angles of attack, claiming accurate reconstruction of unseen configurations and temporal predictions with relative root mean square errors consistently below 2% compared to high-fidelity simulations.

Significance. If the low-rank multilinear separability assumption holds and the reported accuracy is robustly supported, the work provides a mathematically structured, non-intrusive reduced-order modeling approach for multi-parametric dynamical systems in fluid dynamics. By preserving the intrinsic tensor structure while allowing flexible per-mode operator learning, it offers a scalable alternative to black-box surrogates, with built-in capabilities for database completion and resolution enhancement. This could be particularly useful for parametric studies of unsteady flows where full-order simulations are expensive.

major comments (2)
  1. Abstract: The headline claim of relative RMS errors consistently below 2% for reconstruction of unseen Re/AoA configurations and temporal forecasts is presented without any mention of dataset size, the procedure used to select HOSVD truncation ranks, baseline comparisons against other reduced-order or surrogate methods, or statistical error bars. This absence leaves the empirical support for the central accuracy claim provisional and load-bearing for the validation narrative.
  2. Abstract (methodology description): The decoupled operator strategy (GPR on parametric/spatial modes, RNN on temporal mode) rests on the assumption that the unsteady flow data admits a sufficiently low-rank multilinear structure permitting independent approximation without destroying essential nonlinear couplings. In the laminar regime, variations in Reynolds number and angle of attack can produce non-separable effects (e.g., vortex shedding frequency depending simultaneously on both parameters); if this occurs, the independent operators will misrepresent extrapolation cases even when in-sample reconstruction succeeds. No rank-truncation diagnostics, core-tensor analysis, or extrapolation-specific error breakdowns are referenced to confirm the assumption holds at the stated accuracy level.
minor comments (1)
  1. Abstract: The acronym 'MoTIF' is introduced without an explicit expansion or definition of its components in the provided text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below, clarifying aspects of the manuscript and indicating revisions where the abstract or supporting analysis can be strengthened without altering the core claims.

read point-by-point responses

  1. Referee: Abstract: The headline claim of relative RMS errors consistently below 2% for reconstruction of unseen Re/AoA configurations and temporal forecasts is presented without any mention of dataset size, the procedure used to select HOSVD truncation ranks, baseline comparisons against other reduced-order or surrogate methods, or statistical error bars. This absence leaves the empirical support for the central accuracy claim provisional and load-bearing for the validation narrative.

    Authors: We agree that the abstract, as a concise summary, omits several supporting details that appear in the main text. The dataset size, HOSVD rank selection via energy threshold on the singular values, baseline comparisons to POD-based surrogates, and cross-validation error bars are all reported in Sections 3 and 4. To improve clarity for readers who focus on the abstract, we have revised it to include a brief statement on dataset scale and the rank-selection criterion while preserving length constraints. revision: yes

  2. Referee: Abstract (methodology description): The decoupled operator strategy (GPR on parametric/spatial modes, RNN on temporal mode) rests on the assumption that the unsteady flow data admits a sufficiently low-rank multilinear structure permitting independent approximation without destroying essential nonlinear couplings. In the laminar regime, variations in Reynolds number and angle of attack can produce non-separable effects (e.g., vortex shedding frequency depending simultaneously on both parameters); if this occurs, the independent operators will misrepresent extrapolation cases even when in-sample reconstruction succeeds. No rank-truncation diagnostics, core-tensor analysis, or extrapolation-specific error breakdowns are referenced to confirm the assumption holds at the stated accuracy level.

    Authors: We acknowledge the referee's concern about potential non-separability. The manuscript already contains core-tensor analysis and singular-value spectra (Section 2.3 and Figure 3) showing that the leading modes capture the dominant energy, including coupled parametric effects on shedding frequency. The parametric GPR operators are trained on the joint variation of Re and AoA, allowing the model to learn non-separable influences within the low-rank subspace. We have added explicit extrapolation error breakdowns for unseen (Re, AoA) pairs in the revised results section to further substantiate the claim. We also note the limitation for strongly nonlinear regimes in the updated discussion. revision: partial

Circularity Check

0 steps flagged

No circularity: derivation chain remains self-contained

full rationale

The paper decomposes multi-parametric unsteady flow data via HOSVD into distinct modal components for parameters, space, and time, then applies independent GPR operators to parametric/spatial modes and RNN to the temporal mode for interpolation, extrapolation, and forecasting. Validation explicitly targets unseen Reynolds-number and angle-of-attack configurations with reported RMSE <2% against held-out high-fidelity simulations. No equation or step reduces the claimed predictions to quantities fitted on the identical data used for validation; the low-rank multilinear assumption is an empirical modeling choice whose validity is tested externally rather than enforced by construction. The central claims therefore retain independent content beyond the input dataset.

Axiom & Free-Parameter Ledger

3 free parameters · 1 axioms · 0 invented entities

The method rests on the assumption that the flow data tensor admits a useful low-rank multilinear factorization; hyperparameters of the GPR and RNN are fitted but not enumerated.

free parameters (3)
  • HOSVD truncation ranks
    Chosen to retain dominant modes while compressing the data; exact values not stated in abstract.
  • GPR kernel hyperparameters
    Length scales and variance parameters fitted to modal matrices for interpolation and extrapolation.
  • RNN training hyperparameters
    Network depth, hidden size, and learning schedule fitted to the temporal mode.
axioms (1)
  • domain assumption Unsteady flow data can be represented as a low-rank tensor whose modes for parameters, space, and time are separable enough for independent operators.
    Invoked when HOSVD is applied to obtain the structured multilinear representation.

pith-pipeline@v0.9.0 · 5761 in / 1434 out tokens · 38856 ms · 2026-05-18T03:03:50.358181+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

54 extracted references · 54 canonical work pages · 2 internal anchors

  1. [1]

    The impact of oceanic heat transport on the atmospheric circulation,

    M.-A. Knietzsch, V. Lucarini, and F. Lunkeit, “The impact of oceanic heat transport on the atmospheric circulation,”arXiv preprint arXiv:1410.0565, 2014

    work page arXiv 2014

  2. [2]

    Numerical simulation of turbulent sediment transport, from bed load to saltation

    O. Duran, B. Andreotti, and P. Claudin, “Numerical simulation of turbulent sediment transport, from bed load to saltation,”arXiv preprint arXiv:1111.6898, 2011

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  3. [3]

    Urban fluid mechanics: Air circulation and contaminant dispersion in cities,

    E. R. Pardyjak, M. J. Brown, and J. C. Hunt, “Urban fluid mechanics: Air circulation and contaminant dispersion in cities,”Environmental Fluid Mechanics, vol. 2, no. 1, pp. 95–116, 2002

    work page 2002

  4. [4]

    Effect of water–air heat transfer on the spread of thermal pollution in rivers,

    A. Rajwa-Kuligiewicz, K. Kulasza, and A. Kiczko, “Effect of water–air heat transfer on the spread of thermal pollution in rivers,”Acta Geophysica, vol. 67, no. 2, pp. 597–619, 2019

    work page 2019

  5. [5]

    contributors,Application of cfd in thermal power plants,https://en.wikipedia.org/wiki/ Application_of_CFD_in_thermal_power_plants, Accessed: 2025-05-22, 2021

    W. contributors,Application of cfd in thermal power plants,https://en.wikipedia.org/wiki/ Application_of_CFD_in_thermal_power_plants, Accessed: 2025-05-22, 2021

    work page 2025

  6. [6]

    The future of computational fluid dynamics (CFD) simulation in the chemical process industries,

    D. F. Fletcher, “The future of computational fluid dynamics (CFD) simulation in the chemical process industries,”Chemical Engineering Research and Design, vol. 187, pp. 299–305, 2022, issn: 0263-8762.doi:https://doi.org/10.1016/j.cherd.2022.09.021[Online]. Available: https://www.sciencedirect.com/science/article/pii/S0263876222005202

    work page doi:10.1016/j.cherd.2022.09.021 2022

  7. [7]

    Handling simulation failures of a computationally expensive multiobjective optimization problem in pump design,

    A. Mazumdar, J. Burkotov´ a, T. Kr´ atk´ y, T. Chugh, and K. Miettinen, “Handling simulation failures of a computationally expensive multiobjective optimization problem in pump design,” Engineering Applications of Artificial Intelligence, vol. 136, p. 108 897, 2024,issn: 0952-1976. doi:https : / / doi . org / 10 . 1016 / j . engappai . 2024 . 108897[Onlin...

    work page 2024

  8. [8]

    Z. Dar, J. Baiges, and R. Codina,Reduced Order Modeling. Unspecified Publisher, 2024

    work page 2024

  9. [9]

    On the experimental, numerical and data-driven methods to study urban flows,

    P. Torres, S. Le Clainche, and R. Vinuesa, “On the experimental, numerical and data-driven methods to study urban flows,”Energies, vol. 14, no. 5, 2021,issn: 1996-1073.doi:10.3390/ en14051310[Online]. Available:https://www.mdpi.com/1996-1073/14/5/1310

    work page 2021

  10. [10]

    Parametric dynamic mode decomposition for reduced order modeling,

    Q. A. Huhn, M. E. Tano, J. C. Ragusa, and Y. Choi, “Parametric dynamic mode decomposition for reduced order modeling,”Journal of Computational Physics, vol. 475, p. 111 852, 2023,issn: 0021-9991.doi:https://doi.org/10.1016/j.jcp.2022.111852[Online]. Available:https: //www.sciencedirect.com/science/article/pii/S0021999122009159

    work page doi:10.1016/j.jcp.2022.111852 2023

  11. [11]

    A novel tensor-based modal decomposition method for reduced order modeling and optimal sparse sensor placement,

    Z. Zhong, X. Hua, Z. Zhai, and M. Ma, “A novel tensor-based modal decomposition method for reduced order modeling and optimal sparse sensor placement,”Aerospace Science and Tech- nology, vol. 155, p. 109 530, 2024,issn: 1270-9638.doi:https://doi.org/10.1016/j.ast. 2024.109530[Online]. Available:https://www.sciencedirect.com/science/article/pii/ S127096382...

    work page doi:10.1016/j.ast 2024

  12. [12]

    TURBULENCE AND THE DYNAMICS OF COHERENT STRUCTURES PART I: COHERENT STRUCTURES,

    L. Sirovich, “TURBULENCE AND THE DYNAMICS OF COHERENT STRUCTURES PART I: COHERENT STRUCTURES,”Quarterly of Applied Mathematics, vol. 45, no. 3, pp. 561–571, 1987,issn: 0033569X, 15524485. Accessed: Oct. 4, 2024. [Online]. Available:http://www.jstor. org/stable/43637457

    work page arXiv 1987

  13. [13]

    Turbulence and the Dynamics of Coherent Structures. II. Symmetries and Trans- formations,

    L. Sirovich, “Turbulence and the Dynamics of Coherent Structures. II. Symmetries and Trans- formations,”Quarterly of Applied Mathematics, vol. 45, pp. 573–582, 1987. [Online]. Available: https://api.semanticscholar.org/CorpusID:11429443

    work page 1987

  14. [14]

    J. L. Lumley,The structure of inhomogeneous turbulent flows, 1967

    work page 1967

  15. [15]

    Model order reduction in aerodynamics: Review and appli- cations,

    G. Mendon¸ ca, F. Afonso, and F. Lau, “Model order reduction in aerodynamics: Review and appli- cations,”Proceedings of the Institution of Mechanical Engineers, Part G: Journal of Aerospace Engineering, vol. 233, no. 15, pp. 5816–5836, 2019.doi:10.1177/0954410019853472eprint: https://doi.org/10.1177/0954410019853472. [Online]. Available:https://doi.org/10...

    work page doi:10.1177/0954410019853472eprint: 2019

  16. [16]

    A transient reduced order model for battery thermal management based on singular value decomposition,

    X. Hu et al., “A transient reduced order model for battery thermal management based on singular value decomposition,” in2014 IEEE Energy Conversion Congress and Exposition (ECCE), 2014, pp. 3971–3976.doi:10.1109/ECCE.2014.6953941

    work page doi:10.1109/ecce.2014.6953941 2014

  17. [18]

    The extension of factor analysis to three-dimensional matrices,

    L. R. Tucker, “The extension of factor analysis to three-dimensional matrices,” inContributions to mathematical psychology.H. Gulliksen and N. Frederiksen, Eds., New York: Holt, Rinehart and Winston, 1964, pp. 110–127

    work page 1964

  18. [19]

    Quaternion higher-order singular value decomposition and its applications in color image processing,

    J. Miao, K. I. Kou, D. Cheng, and W. Liu, “Quaternion higher-order singular value decomposition and its applications in color image processing,”Information Fusion, vol. 92, pp. 139–153, 2023, issn: 1566-2535.doi:https://doi.org/10.1016/j.inffus.2022.11.026[Online]. Available: https://www.sciencedirect.com/science/article/pii/S1566253522002366

    work page doi:10.1016/j.inffus.2022.11.026 2023

  19. [20]

    Nonintrusive model order reduction for cross- diffusion systems,

    B. Karas¨ ozen, G. M¨ ulayim, and M. Uzunca, “Nonintrusive model order reduction for cross- diffusion systems,”Communications in Nonlinear Science and Numerical Simulation, vol. 115, p. 106 734, 2022,issn: 1007-5704.doi:https://doi.org/10.1016/j.cnsns.2022.106734[On- line]. Available:https://www.sciencedirect.com/science/article/pii/S1007570422002738

    work page doi:10.1016/j.cnsns.2022.106734 2022

  20. [21]

    A high fidelity cost efficient tensorial method based on combined pod-hosvd reduced order model of flow field,

    M. K. Moayyedi and M. N. beygi, “A high fidelity cost efficient tensorial method based on combined pod-hosvd reduced order model of flow field,”European Journal of Computational Mechanics, vol. 27, no. 4, pp. 342–366, 2018.doi:10.1080/17797179.2018.1550963eprint: https://doi.org/10.1080/17797179.2018.1550963. [Online]. Available:https://doi.org/ 10.1080/1...

    work page doi:10.1080/17797179.2018.1550963eprint: 2018

  21. [22]

    Application of hosvd to aerodynamics. the problem of shock waves,

    L. S. Lorente, D. Alonso, J. M. Vega de Prada, and A. Vel´ azquez, “Application of hosvd to aerodynamics. the problem of shock waves,” in9th International Conference on Computational and Mathematical Methods in Science and Engineering, CMMSE 2009, Asturias, Espa˜ na, Jun. 2009,isbn: 978-84-612-9727-6

    work page 2009

  22. [23]

    Deep learning com- bined with singular value decomposition to reconstruct databases in fluid dynamics[formula pre- sented],

    P. D´ ıaz-Morales, A. Corrochano, M. L´ opez-Mart´ ın, and S. L. Clainche, “Deep learning com- bined with singular value decomposition to reconstruct databases in fluid dynamics[formula pre- sented],”Expert Systems with Applications, vol. 238, Mar. 2024.doi:10.1016/j.eswa.2023. 121924

    work page doi:10.1016/j.eswa.2023 2024

  23. [24]

    Data repairing and resolution enhancement using data-driven modal decomposition and deep learning,

    A. Hetherington, D. Serfaty, A. Corrochano, J. Soria, and S. Le Clainche, “Data repairing and resolution enhancement using data-driven modal decomposition and deep learning,”Experi- mental Thermal and Fluid Science, vol. 157, p. 111 241, 2024,issn: 0894-1777.doi:https : / / doi . org / 10 . 1016 / j . expthermflusci . 2024 . 111241[Online]. Available:http...

    work page 2024

  24. [25]

    Digital twin of a combustion furnace operating in flameless conditions: Reduced-order model development from cfd simulations,

    G. Aversano, M. Ferrarotti, and A. Parente, “Digital twin of a combustion furnace operating in flameless conditions: Reduced-order model development from cfd simulations,”Proceedings of the Combustion Institute, vol. 38, pp. 5373–5381, 2021.doi:10.1016/j.proci.2020.08.032 [Online]. Available:https://doi.org/10.1016/j.proci.2020.08.032

    work page doi:10.1016/j.proci.2020.08.032 2021

  25. [26]

    Adaptive digital twins of combustion systems using sparse sensing strategies,

    A. Procacci, R. Amaduzzi, A. Coussement, and A. Parente, “Adaptive digital twins of combustion systems using sparse sensing strategies,”Proceedings of the Combustion Institute, 2022.doi: 10.1016/j.proci.2022.03.158[Online]. Available:https://doi.org/10.1016/j.proci. 2022.03.158

    work page doi:10.1016/j.proci.2022.03.158 2022

  26. [27]

    A predictive hybrid reduced order model based on proper orthogonal decomposition combined with deep learning architectures,

    R. Abad´ ıa-Heredia, M. L´ opez-Mart´ ın, B. Carro, J. Arribas, J. P´ erez, and S. Le Clainche, “A predictive hybrid reduced order model based on proper orthogonal decomposition combined with deep learning architectures,”Expert Systems with Applications, vol. 187, p. 115 910, 2022, issn: 0957-4174.doi:https://doi.org/10.1016/j.eswa.2021.115910[Online]. Av...

    work page doi:10.1016/j.eswa.2021.115910 2022

  27. [28]

    Pod-lstm model for predicting pressure time series on structures,

    X. Du, C. Hu, and H. Dong, “Pod-lstm model for predicting pressure time series on structures,” Journal of Wind Engineering and Industrial Aerodynamics, vol. 245, p. 105 651, 2024,issn: 0167-6105.doi:https : / / doi . org / 10 . 1016 / j . jweia . 2024 . 105651[Online]. Available: https://www.sciencedirect.com/science/article/pii/S016761052400014X

    work page 2024

  28. [29]

    Generation of aerodynamics databases using high- order singular value decomposition,

    L. S. Lorente, J. M. Vega, and A. Velazquez, “Generation of aerodynamics databases using high- order singular value decomposition,”Journal of Aircraft, vol. 45, no. 5, pp. 1779–1788, 2008. doi:10 . 2514 / 1 . 35258eprint:https : / / doi . org / 10 . 2514 / 1 . 35258. [Online]. Available: https://doi.org/10.2514/1.35258

    work page doi:10.2514/1.35258 2008

  29. [30]

    Barrag´ an, A

    G. Barrag´ an, A. Hetherington, R. Abad´ ıa-Heredia, J. Garicano-Mena, and S. L. Clainche,Hosvd- sr: A physics-based deep learning framework for super-resolution in fluid dynamics, 2025. arXiv: 2504.17994 [physics.flu-dyn]. [Online]. Available:https://arxiv.org/abs/2504.17994

    work page arXiv 2025

  30. [31]

    Sengupta, R

    A. Sengupta, R. Abad´ ıa-Heredia, A. Hetherington, J. M. P´ erez, and S. L. Clainche,Hybrid machine learning models based on physical patterns to accelerate cfd simulations: A short guide on autoregressive models, 2025. arXiv:2504 . 06774 [physics.flu-dyn]. [Online]. Available: https://arxiv.org/abs/2504.06774

    work page arXiv 2025

  31. [32]

    Importance of input data normalization for the application of neural net- works to complex industrial problems,

    J. Sola and J. Sevilla, “Importance of input data normalization for the application of neural net- works to complex industrial problems,”IEEE Transactions on Nuclear Science, vol. 44, pp. 1464– 1468, 3 PART 3 1997,issn: 00189499.doi:10.1109/23.589532

    work page doi:10.1109/23.589532 1997

  32. [33]

    Investigating the impact of min-max data normalization on the regression perfor- mance of k-nearest neighbor with different similarity measurements,

    P. J. M. Ali, “Investigating the impact of min-max data normalization on the regression perfor- mance of k-nearest neighbor with different similarity measurements,”ARO-THE SCIENTIFIC JOURNAL OF KOYA UNIVERSITY, vol. 10, pp. 85–91, 1 Jun. 2022,issn: 2410-9355.doi: 10.14500/aro.10955

    work page doi:10.14500/aro.10955 2022

  33. [34]

    Normalization techniques in training dnns: Methodology, analysis and application,

    L. Huang, J. Qin, Y. Zhou, F. Zhu, L. Liu, and L. Shao, “Normalization techniques in training dnns: Methodology, analysis and application,” Sep. 2020. [Online]. Available:http://arxiv. org/abs/2009.12836

    work page arXiv 2020

  34. [35]

    Implications of factor analysis of three-way matrices for measurement of change,

    L. R. Tucker, “Implications of factor analysis of three-way matrices for measurement of change,” inProblems in measuring change.C. W. Harris, Ed., Madison WI: University of Wisconsin Press, 1963, pp. 122–137

    work page 1963

  35. [36]

    Some mathematical notes on three-mode factor analysis,

    L. R. Tucker, “Some mathematical notes on three-mode factor analysis,”Psychometrika, vol. 31, pp. 279–311, 1966. [Online]. Available:https : / / api . semanticscholar . org / CorpusID : 44301099

    work page 1966

  36. [37]

    On the best rank-1 and rank-(r1 ,r2 ,. . .,rn) approximation of higher-order tensors,

    L. De Lathauwer, B. De Moor, and J. Vandewalle, “On the best rank-1 and rank-(r1 ,r2 ,. . .,rn) approximation of higher-order tensors,”SIAM Journal on Matrix Analysis and Applica- tions, vol. 21, no. 4, pp. 1324–1342, 2000.doi:10.1137/S0895479898346995eprint:https: //doi.org/10.1137/S0895479898346995. [Online]. Available:https://doi.org/10.1137/ S0895479898346995

    work page doi:10.1137/s0895479898346995eprint:https: 2000

  37. [38]

    SIAM Journal on Matrix Analysis and Applications , author =

    L. De Lathauwer, B. De Moor, and J. Vandewalle, “A multilinear singular value decomposition,” SIAM Journal on Matrix Analysis and Applications, vol. 21, no. 4, pp. 1253–1278, 2000.doi:10. 1137/S0895479896305696eprint:https://doi.org/10.1137/S0895479896305696. [Online]. Available:https://doi.org/10.1137/S0895479896305696 34

    work page doi:10.1137/s0895479896305696 2000

  38. [39]

    Compression of aerodynamic databases using high-order singular value decomposition,

    L. S. Lorente, J. M. Vega, and A. Velazquez, “Compression of aerodynamic databases using high-order singular value decomposition,”Aerospace Science and Technology, vol. 14, no. 3, pp. 168–177, 2010,issn: 1270-9638.doi:10 . 1016 / j . ast . 2009 . 12 . 003[Online]. Available: https://www.sciencedirect.com/science/article/pii/S1270963809000935

    work page 2010

  39. [40]

    Modelflows-app: Data-driven post-processing and reduced order mod- elling tools,

    A. Hetherington et al., “Modelflows-app: Data-driven post-processing and reduced order mod- elling tools,”Computer Physics Communications, vol. 301, p. 109 217, 2024,issn: 0010-4655. doi:https : / / doi . org / 10 . 1016 / j . cpc . 2024 . 109217[Online]. Available:https : / / www . sciencedirect.com/science/article/pii/S0010465524001401

    work page 2024

  40. [41]

    Aaron Towne, Oliver T

    K. Taira et al., “Modal analysis of fluid flows: An overview,”AIAA Journal, vol. 55, no. 12, pp. 4013–4041, 2017.doi:10.2514/1.J056060

    work page doi:10.2514/1.j056060 2017

  41. [42]

    Holmes, J

    P. Holmes, J. L. Lumley, G. Berkooz, and C. W. Rowley,Turbulence, Coherent Structures, Dynamical Systems and Symmetry. Cambridge University Press, 2012

    work page 2012

  42. [43]

    A. N. Kolmogorov,“On the analytic methods of probability theory. Uspekhi matematicheskikh nauk, 1938

    work page 1938

  43. [44]

    I. A. Ibragimov and Y. A. Rozanov,Gaussian random processes. Springer Science & Business Media, 1978

    work page 1978

  44. [45]

    C. E. Rasmussen and C. K. I. Williams,Gaussian Processes for Machine Learning. The MIT Press, 2006

    work page 2006

  45. [46]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural Computation, vol. 9, no. 8, pp. 1735–1780, Nov. 1997,issn: 0899-7667.doi:10.1162/neco.1997.9.8.1735eprint: https://direct.mit.edu/neco/article-pdf/9/8/1735/813796/neco.1997.9.8.1735.pdf. [Online]. Available:https://doi.org/10.1162/neco.1997.9.8.1735

    work page doi:10.1162/neco.1997.9.8.1735eprint: 1997

  46. [47]

    A survey on long short-term memory networks for time series prediction,

    B. Lindemann, T. M¨ uller, H. Vietz, N. Jazdi, and M. Weyrich, “A survey on long short-term memory networks for time series prediction,”Procedia CIRP, vol. 99, pp. 650–655, 2021, 14th CIRP Conference on Intelligent Computation in Manufacturing Engineering, 15-17 July 2020, issn: 2212-8271.doi:https://doi.org/10.1016/j.procir.2021.03.088[Online]. Available...

    work page doi:10.1016/j.procir.2021.03.088 2021

  47. [48]

    An lstm-enhanced surrogate model to simulate the dynamics of particle-laden fluid systems,

    A. Hajisharifi, R. Halder, M. Girfoglio, A. Beccari, D. Bonanni, and G. Rozza, “An lstm-enhanced surrogate model to simulate the dynamics of particle-laden fluid systems,”Computers & Fluids, vol. 280, p. 106 361, 2024,issn: 0045-7930.doi:https://doi.org/10.1016/j.compfluid. 2024.106361[Online]. Available:https://www.sciencedirect.com/science/article/pii/ ...

    work page doi:10.1016/j.compfluid 2024

  48. [49]

    Chollet et al.,Keras,https://github.com/fchollet/keras, 2015

    F. Chollet et al.,Keras,https://github.com/fchollet/keras, 2015

    work page 2015

  49. [50]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural computation, vol. 9, no. 8, pp. 1735–1780, 1997

    work page 1997

  50. [51]

    D. P. Kingma and J. Ba,Adam: A method for stochastic optimization, 2017. arXiv:1412.6980 [cs.LG]. [Online]. Available:https://arxiv.org/abs/1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  51. [52]

    Turbulent flow around a square cylinder at reynolds number 22,000: A dns study,

    F. Trias, A. Gorobets, and A. Oliva, “Turbulent flow around a square cylinder at reynolds number 22,000: A dns study,”Computers & Fluids, vol. 123, pp. 87–98, 2015,issn: 0045-7930. doi:https : / / doi . org / 10 . 1016 / j . compfluid . 2015 . 09 . 013[Online]. Available:https : //www.sciencedirect.com/science/article/pii/S0045793015003254

    work page 2015

  52. [53]

    Wake interactions in multibody con- figurations with different shape,

    V. Beltr´ an, C. M´ endez, S. Le Clainche, and J. M. Vega, “Wake interactions in multibody con- figurations with different shape,”Aerospace Science and Technology, vol. 95, p. 105 428, 2019, issn: 1270-9638.doi:https://doi.org/10.1016/j.ast.2019.105428[Online]. Available: https://www.sciencedirect.com/science/article/pii/S1270963819311307

    work page doi:10.1016/j.ast.2019.105428 2019

  53. [54]

    Study of various mesh strategies for flow over a square cylinder using OpenFOAM®,

    A. S. Kumar, A. Singh, N. Gaur, and K. B. T., “Study of various mesh strategies for flow over a square cylinder using OpenFOAM®,”AIP Conference Proceedings, vol. 2277, no. 1, p. 230 001, Nov. 2020,issn: 0094-243X.doi:10.1063/5.0025201eprint:https://pubs.aip.org/aip/ acp/article- pdf/doi/10.1063/5.0025201/13272208/230001\_1\_online.pdf. [Online]. Available...

    work page doi:10.1063/5.0025201eprint:https://pubs.aip.org/aip/ 2020

  54. [55]

    OpenFOAM: Open source CFD in research and industry,

    H. Jasak, “OpenFOAM: Open source CFD in research and industry,”International Journal of Naval Architecture and Ocean Engineering, vol. 1, no. 2, pp. 89–94, 2009,issn: 2092-6782. doi:https : / / doi . org / 10 . 2478 / IJNAOE - 2013 - 0011[Online]. Available:https : / / www . sciencedirect.com/science/article/pii/S2092678216303879 36

    work page 2009