Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

Machine Learning-based quadratic closures for non-intrusive Reduced Order Models

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read A neural-operator-parameterized quadratic correction re-introduces neglected POD modes and reduces prediction error by up to 90% over the linear ROM baseline on two fluid benchmarks.

desk verdict Useful moderate extension of quadratic closures; the 80% claim over Quad-LS rests on an underdetermined baseline, and discretization agnosticism is untested. read the letter →

arxiv 2506.09830 v2 pith:KBW6C5VR submitted 2025-06-11 math.NA cs.NA

classification math.NAcs.NA MSC 68T0735Q3565N99
keywords ReducedOrderModelsProperOrthogonalDecompositionQuadraticclosureNeuraloperatorsMIONetDeepONetNon-intrusiveROMFluiddynamics
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

This paper tries to show that a quadratic closure for POD-based non-intrusive reduced-order models can be made substantially more accurate by learning the quadratic operator with a neural operator rather than fitting one global tensor by least squares. The proposed models, QuadNet and QuadNet-$\mu$, add a correction $a(\mu)^T C(\Phi_r,x,\mu)a(\mu)$ to the linear POD reconstruction, where $C$ is produced by a DeepONet or MIONet and is therefore continuous in space and, in the $\mu$ version, in the parameter. On two fluid benchmarks the corrected models cut test errors by up to 90 percent relative to the linear POD-RBF baseline and up to 80 percent relative to the least-squares quadratic closure. The method matters because it offers a route to more accurate ROMs in the under-resolved, scarce-data regime while preserving the interpretable POD-plus-quadratic structure.

What carries the argument

The load-bearing object is the quadratic closure operator $C(\Phi_r,x,\mu)$, which replaces the fixed tensor of the least-squares closure with the output of a neural operator. In QuadNet the operator is built with a DeepONet-like split, a branch network that reads the $r$ POD modes and a trunk network that reads spatial coordinates, followed by a small combining network that outputs the symmetric quadratic form; in QuadNet-$\mu$ a second branch network also reads the parameter $\mu$, giving a MIONet structure, a multi-input operator network with separate branch networks for each input type. This makes the correction continuous in space and parameter, so the trained operator can be evaluated at any mesh point and adapts locally in parameter space, while training with a relative-error loss keeps small-magnitude corrections from being swamped by scale differences.

What would settle it

Manufacture a test problem whose exact correction is cubic, such as $\tau_{\mathrm{exact}}=(a^T a)a$ scaled by a known field, train QuadNet-$\mu$ on abundant data, and check whether the learned correction can reduce the residual below the best quadratic approximation; if it cannot, the quadratic ansatz, not the network, is the limiting factor.

Watch

Extended reading notes

Core claim

The central claim is that the exact truncation error $\tau_{\mathrm{exact}}(\mu)=u(\hat{x},\mu)-\sum_i a_i(\mu)\phi_i(\hat{x})$ can be approximated by a quadratic form $a(\mu)^T C(\Phi_r,x,\mu)a(\mu)$ in the retained POD coefficients, with the operator $C$ emitted by a multi-input operator network. Because the network receives the POD modes, the spatial coordinate, and the parameter as inputs, the correction varies continuously in space and in parameter space, so it is independent of the discretization and acts as a local closure rather than one global least-squares tensor. In the backward-facing step and lid-driven cavity benchmarks, the parameter-dependent version, QuadNet-$\mu$, reports the lowest mean, median, and spread of errors, with the paper summarizing gains of up to 90 percent over POD-RBF and up to 80 percent over Quad-LS, including when training uses only a subset of spatial points or a small number of snapshots.

Load-bearing premise

The load-bearing premise is that the entire neglected part of the flow can be written as quadratic combinations of the retained POD coefficients; if the missing part depends on those coefficients in a cubic or higher-order way, the correction is misspecified and the reported gains may not transfer to harder problems.

Editorial extensions

If this is right

  • If the central claim holds, the neural-operator closure is the more reliable correction in the under-resolved regime: QuadNet-$\mu$ beats both POD-RBF and Quad-LS on the two tested flow problems, with the largest relative gains when few POD modes are kept.
  • Because the operator is space-continuous, a model trained on 10 to 50 percent of the mesh nodes keeps accuracy close to full-mesh training, lowering offline memory and training cost.
  • In scarce-data settings the neural closure is more robust than the least-squares tensor, which can produce errors larger than the uncorrected linear model when snapshots are few.
  • The same architecture can be adjusted to higher-order corrections by changing the network's output, so the framework is not limited to quadratic forms.

Reading between the lines

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

  • Beyond the paper, the quadratic ansatz is the real ceiling: on problems where the neglected-mode contribution has significant cubic or higher-order dependence on $a$, the same network would need a higher-degree output, and the reported gains should be expected to shrink.
  • Because $C$ is evaluated pointwise in space, a trained QuadNet-$\mu$ could plausibly be queried on a different or re-meshed domain without retraining; the paper demonstrates training on partial meshes but not cross-mesh transfer, so this is a testable extrapolation.
  • The Boltzmann-like collocation-point sampling could be turned into an active-learning loop that adds points where the current network's residual is largest; the paper introduces the sampling idea but does not explore that adaptive strategy.
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

4 major / 5 minor

Summary. The paper proposes a data-driven quadratic closure for non-intrusive ROMs built on POD. The correction term tau(x;mu) = a^T C(Phi_r,x,mu) a is added to the POD reconstruction, and the operator C is learned either by least squares (Quad-LS), by a DeepONet (QuadNet), or by a MIONet with parameter and space inputs (QuadNet-mu). The methods are tested on the backward-facing step and lid-driven cavity benchmarks, with fixed-mode comparisons, partial-observation experiments, and a scarce-data study over (r, N_mu) combinations. The central empirical claims are that QuadNet-mu improves over POD-RBF by up to 90% and over Quad-LS by up to 80% in relative error.

Significance. If the numerical claims are robust, the paper makes a useful contribution: the parameter- and space-continuous parametrization of the quadratic operator is a natural extension of quadratic-manifold ROMs, and the reported gains in the scarce-data regime would be practically relevant. The paper also gives a clear presentation of the loss function, the architecture, and the test cases, and the fixed-mode comparison in Section 3.4 is well-posed. The main weakness is that the headline comparison against Quad-LS relies on an unregularized, underdetermined baseline, and the neural results are reported without repeated training runs. These issues must be addressed before the central claims can be accepted as stated.

major comments (4)
  1. [Section 3.6, Eqs. (7) and (11), Figures 8-9] The Quad-LS baseline in the scarce-data comparison is unregularized. For r=7 (S=28) and r=9 (S=45), with N_mu=10,20,30,40, the per-node least-squares system has more unknowns than equations; the paper itself states that Quad-LS 'struggles in providing meaningful corrections' in this regime. The conclusion's 'up to 80% error reduction compared to Quad-LS' is therefore computed against a degenerate baseline. Please add a regularized Quad-LS variant (e.g., ridge or truncated SVD) and a Quad-LS trained with the same relative-error weighting as in Eq. (13), and recompute the reported improvements.
  2. [Tables 3-4 and Figures 8-9] All numerical results are reported for a single training run per configuration. Since QuadNet and QuadNet-mu are stochastic neural networks, and Section 3.6 acknowledges sensitivity to weight initialization and hyperparameters, the reported rankings may be seed artifacts. Please provide means and standard deviations over at least several random seeds, and state whether the observed differences are statistically significant.
  3. [Section 3.5 and abstract] The claim that the model is 'agnostic of the domain discretisation' is not supported by the experiments. Training on 10%, 20%, or 50% of the nodes of the same mesh and evaluating on the remaining nodes of that same mesh tests spatial subsampling, not independence from the discretization. Please either test on a genuinely different mesh or off-mesh evaluation points, or soften the claim to 'space-continuous' rather than discretization-agnostic.
  4. [Eq. (14)] The exact correction is assumed to be quadratic in the retained POD coefficients, but the manuscript does not provide any diagnostic of how well this ansatz represents tau_exact on the training data. Please report the relative residual of the best quadratic fit or an analysis of the leading omitted terms, so that the reader can assess the misspecification risk when the method is applied to flows with stronger nonlinearity.
minor comments (5)
  1. [Tables 1-2] The 'Min. loss 10^-2' column is unclear; please specify whether it is a stopping criterion, a target loss, or a lower bound for the learning rate.
  2. [Throughout] There are several typographical and formatting issues: 'F unding' in the funding line, 'Sentries' for 'S entries', inconsistent spacing in 'QuadNet -mu', and 'Ntrain_mu' spacing. Please proofread.
  3. [Eq. (19)] The normalization parameter tilde_tau is not precisely defined; please specify whether it is the mean or maximum of the correction norm over nodes or snapshots.
  4. [Section 3.4] The statement that 'QuadNet is not necessarily better than Quad-LS' is an understatement given Tables 3 and 4, where QuadNet is consistently worse; please rephrase to reflect the observed ranking.
  5. [Section 3.5] The reported approximate 10% training-time reduction is based on runs on a non-dedicated workstation; please report more robust timing statistics or acknowledge the noise in these measurements.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the neural closure operators are trained on source snapshots and evaluated on held-out configurations, and the quadratic ansatz is an explicit modeling assumption rather than a derived result.

full rationale

The paper's central derivation is a data-driven correction pipeline: the exact correction is defined in Eq. (5), a quadratic ansatz is introduced in Eqs. (6), (12), and (14), and the operator C is learned by minimizing the relative-error loss (13) on training snapshots. Predictions are then evaluated on held-out test snapshots in Tables 3-4 and Figures 8-9, so the reported errors are genuine out-of-sample quantities and no fitted parameter is relabeled as a prediction. The quadratic form a^T C(Phi_r, x, mu) a is explicitly presented as an ansatz ('all providing a quadratic ansatz in the reduced coefficients'), not as a theorem derived from Navier-Stokes, so adopting it is an assumption rather than a circular step. Self-citations such as Ref. [16] (DeepONet residual learning) and Refs. [22-23] (closure strategies) are used only as background pointers and do not carry the argument. The skeptical concern that the Quad-LS baseline is underdetermined for some (r, N_mu) combinations in Section 3.6 is an experimental-fairness and correctness question, not a circularity, because the comparison is between two independently trained models on test data. Therefore the derivation chain is self-contained with respect to the paper's claims, and no circular step can be exhibited from the text and equations.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The method is supervised learning: it does not introduce new physical entities. The core assumptions are the quadratic ansatz for the closure, the smoothness of the operator, the representativeness of the training manifold, and the accuracy of the RBF coefficient interpolation. These are stated explicitly in Sections 2 and 3.

free parameters (5)
  • r (number of retained POD modes) = 3 in main setting; 3, 5, 7, 9 in Section 3.6
    User-selected truncation that defines the under-resolved regime; all results depend on it.
  • RBF kernel type = linear for backstep; thin plate spline for cavity
    Chosen per test case in Section 3.3.1; affects baseline POD-RBF accuracy and the coefficients a(mu) used by all corrected models.
  • Network hyperparameters = p=S=6, hidden layers 7x20, Tanh, lr O(10^-3), max epochs 20000, min loss 10^-2
    Hand-tuned and reported as sensitive; Tables 1 and 2 give ranges rather than exact values, so the reported accuracy is conditional on these choices.
  • Training snapshot count N_mu = 400 (backstep), 100 (cavity)
    Fixed for the main comparison; varied in scarce-data experiments. Not fitted, but determines the amount of information the closure can learn.
  • Boltzmann sampling constants = tilde tau and epsilon unspecified
    Partial-data results in Section 3.5 depend on the sampling distribution in Equation (19); the normalization and epsilon are not given, which impacts reproducibility.
assumptions (5)
  • domain assumption The exact residual tau_exact = u minus the POD reconstruction is well approximated by a quadratic form a^T C(Phi_r, x, mu) a in the retained POD coefficients.
    Core ansatz in Equations (12) and (14), inherited from quadratic manifold literature; the paper does not test whether higher-order terms are needed.
  • domain assumption The map C is smooth and learnable by a fully connected DeepONet/MIONet so that it generalizes in space and in the parameter.
    Sections 2.2.2 and 2.2.3 assume neural approximation; no approximation guarantee or monotonicity condition is shown for this specific operator.
  • domain assumption POD modes from training snapshots represent the solution manifold for unseen parameters.
    Section 2.1 builds the basis from the snapshot matrix S; Section 3.6 admits that few snapshots may be non-representative, showing this assumption is load-bearing.
  • domain assumption RBF interpolation yields accurate reduced coefficients a(mu*) at test parameters.
    Equation (2) supplies all a values used in the corrections; its error is never separated from the closure error.
  • standard math FEniCS and OpenFOAM solutions are reliable ground truth.
    Sections 3.1 and 3.2 use these high-fidelity discretizations as reference; convergence or mesh-independence studies are not reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Machine Learning-based quadratic closures for non-intrusive Reduced Order Models." pith.science (2026). https://pith.science/paper/KBW6C5VR

@misc{pith2026250609830,
  author       = {Pith},
  title        = {Pith review of: Machine Learning-based quadratic closures for non-intrusive Reduced Order Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KBW6C5VR}},
  note         = {Machine review of arXiv:2506.09830}
}
read the original abstract

In the present work, we introduce a data-driven approach to enhance the accuracy of non-intrusive Reduced Order Models (ROMs). In particular, we focus on ROMs built using Proper Orthogonal Decomposition (POD) in an under-resolved and marginally-resolved regime, i.e. when the number of modes employed is not enough to capture the system dynamics. We propose a method to re-introduce the contribution of neglected modes through a quadratic correction term, given by the action of a quadratic operator on the POD coefficients. Differently from the state-of-the-art methodologies, where the operator is learned via least-squares optimisation, we propose to parametrise the operator by a Multi-Input Operators Network (MIONet). This way, we are able to build models with higher generalisation capabilities, where the operator itself is continuous in space -- thus agnostic of the domain discretisation -- and parameter-dependent. We test our model on two standard benchmarks in fluid dynamics and show that the correction term improves the accuracy of standard POD-based ROMs.

Figures

Figures reproduced from arXiv: 2506.09830 by the authors.

Figure 1
Figure 1. Schematic representation of the DeepONet architecture employed in QuadNet. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Schematic representation of the MIONet architecture employed in QuadNet - [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Domains and corresponding offline grid for the two test cases. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Decay of the snapshot matrix’s singular values for the two test cases. Values are [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Plots for a random test FOM solution and corresponding error fields for POD-RBF, [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: Comparison between samples of points drawn from different distribution. The blue [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: Boxplots comparing the accuracy of models. [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Comparison of QuadNet -µ errors with POD-RBF and Quad-LS. The left half of the circles shows the relative difference in error between QuadNet -µ and POD-RBF, while the right half shows the difference between QuadNet -µ and Quad-LS. We find that QuadNet -µ performs bett…
Figure 9
Figure 9. Figure 9: Comparison between POD-RBF, Quad-LS and QuadNet - [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Surrogate modeling for convection-dominated parametric problems based on error learning

    math.DS 2026-05 unverdicted novelty 6.0 of 10

    A hybrid projection-ROM plus DNN error-correction surrogate, with a non-intrusive variant, delivers higher accuracy and lower online cost than existing methods on parametric inviscid Burgers equations.

  2. A Quadratic Order Reduction -- Gaussian Process Ordinary Differential Equation framework for the inference of Large Continuous Dynamical Systems

    math.NA 2026-06 unverdicted novelty 5.0 of 10

    A framework merges Gaussian Process ODEs with quadratic reduced-order modeling for stable short-term forecasting of large dynamical systems with uncertainty quantification.

Reference graph

Works this paper leans on

47 extracted references · 41 canonical work pages · cited by 2 Pith papers

  1. [1]

    Radial basis function and related models: an overview

    Felipe Miguel Aparicio Acosta. Radial basis function and related models: an overview. Signal Processing, 45(1):37–58, 1995

  2. [2]

    On closures for reduced order models—a spectrum of first-principle to machine- learned avenues.Physics of Fluids, 33(9), 2021

    Shady E Ahmed, Suraj Pawar, Omer San, Adil Rasheed, Traian Iliescu, and Bernd R Noack. On closures for reduced order models—a spectrum of first-principle to machine- learned avenues.Physics of Fluids, 33(9), 2021. 17

  3. [3]

    Galerkin neural networks: A framework for approxi- mating variational equations with error control.SIAM Journal on Scientific Computing, 43(4):A2474–A2501, 2021

    Mark Ainsworth and Justin Dong. Galerkin neural networks: A framework for approxi- mating variational equations with error control.SIAM Journal on Scientific Computing, 43(4):A2474–A2501, 2021

  4. [4]

    Nonlinear model order reduction via dynamic mode decomposition.SIAM Journal on Scientific Computing, 39(5):B778–B796, 2017

    Alessandro Alla and J Nathan Kutz. Nonlinear model order reduction via dynamic mode decomposition.SIAM Journal on Scientific Computing, 39(5):B778–B796, 2017

  5. [5]

    Nonlinear model order reduc- tion based on local reduced-order bases.International Journal for Numerical Methods in Engineering, 92(10):891–916, 2012

    David Amsallem, Matthew J Zahr, and Charbel Farhat. Nonlinear model order reduc- tion based on local reduced-order bases.International Journal for Numerical Methods in Engineering, 92(10):891–916, 2012

  6. [6]

    Baratta, Joseph P

    Igor A. Baratta, Joseph P. Dean, Jørgen S. Dokken, Michal Habera, Jack S. Hale, Chris N. Richardson, Marie E. Rognes, Matthew W. Scroggs, Nathan Sime, and Garth N. Wells. DOLFINx: The next generation FEniCS problem solving environment, December 2023

  7. [7]

    Joshua Barnett and Charbel Farhat. Quadratic approximation manifold for mitigating the Kolmogorov barrier in nonlinear projection-based model order reduction.Journal of Computational Physics, 464:111348, September 2022

  8. [8]

    De Gruyter, 2021

    Peter Benner, Stefano Grivet-Talocia, Alfio Quarteroni, Gianluigi Rozza, Wil Schilders, and Lu ´ ıs Miguel Silveira.System-and data-driven methods and algorithms. De Gruyter, 2021

Show all 47 references
  1. [9]

    De Gruyter, 2020

    Peter Benner, Wil Schilders, Stefano Grivet-Talocia, Alfio Quarteroni, Gianluigi Rozza, and Lu ´ ıs Miguel Silveira.Model Order Reduction: Volume 2: Snapshot-Based Methods and Algorithms. De Gruyter, 2020

  2. [10]

    De Gruyter, 2020

    Peter Benner, Wil Schilders, Stefano Grivet-Talocia, Alfio Quarteroni, Gianluigi Rozza, and Lu ´ ıs Miguel Silveira.Model order reduction: Volume 3: Applications. De Gruyter, 2020

  3. [11]

    The proper orthogonal decomposition in the analysis of turbulent flows.Annual review of fluid mechanics, 25(1):539–575, 1993

    Gal Berkooz, Philip Holmes, and John L Lumley. The proper orthogonal decomposition in the analysis of turbulent flows.Annual review of fluid mechanics, 25(1):539–575, 1993

  4. [12]

    Radial basis functions.Acta numerica, 9:1–38, 2000

    Martin Dietrich Buhmann. Radial basis functions.Acta numerica, 9:1–38, 2000

  5. [13]

    Physics-informed ma- chine learning for reduced-order modeling of nonlinear problems.Journal of computational physics, 446:110666, 2021

    Wenqian Chen, Qian Wang, Jan S Hesthaven, and Chuhua Zhang. Physics-informed ma- chine learning for reduced-order modeling of nonlinear problems.Journal of computational physics, 446:110666, 2021

  6. [14]

    Gaussian processes for ordinal regression.Journal of machine learning research, 6(7), 2005

    Wei Chu, Zoubin Ghahramani, and Christopher KI Williams. Gaussian processes for ordinal regression.Journal of machine learning research, 6(7), 2005

  7. [15]

    Generative adversarial reduced order modelling.Scientific Reports, 14(1):3826, 2024

    Dario Coscia, Nicola Demo, and Gianluigi Rozza. Generative adversarial reduced order modelling.Scientific Reports, 14(1):3826, 2024

  8. [16]

    A deeponet multi-fidelity approach for residual learning in reduced order modeling.Advanced Modeling and Simulation in Engineering Sciences, 10(1):12, 2023

    Nicola Demo, Marco Tezzele, and Gianluigi Rozza. A deeponet multi-fidelity approach for residual learning in reduced order modeling.Advanced Modeling and Simulation in Engineering Sciences, 10(1):12, 2023

  9. [17]

    Greedy algorithms for reduced bases in banach spaces.Constructive Approximation, 37(3):455–466, Jun 2013

    Ronald DeVore, Guergana Petrova, and Przemyslaw Wojtaszczyk. Greedy algorithms for reduced bases in banach spaces.Constructive Approximation, 37(3):455–466, Jun 2013

  10. [18]

    Operator inference for non-intrusive model reduction with quadratic manifolds.Computer Methods in Applied Mechanics and Engineering, 403:115717, 2023

    Rudy Geelen, Stephen Wright, and Karen Willcox. Operator inference for non-intrusive model reduction with quadratic manifolds.Computer Methods in Applied Mechanics and Engineering, 403:115717, 2023

  11. [19]

    Singular value decomposition and least squares solutions

    Gene H Golub and Christian Reinsch. Singular value decomposition and least squares solutions. InHandbook for Automatic Computation: Volume II: Linear Algebra, pages 134–151. Springer, 1971

  12. [20]

    Non-intrusive reduced order modeling of nonlinear problems using neural networks.Journal of Computational Physics, 363, 02 2018

    Jan Hesthaven and Stefano Ubbiali. Non-intrusive reduced order modeling of nonlinear problems using neural networks.Journal of Computational Physics, 363, 02 2018. 18

  13. [21]

    Saddam Hijazi, Melina Freitag, and Niels Landwehr. POD-Galerkin reduced order models and physics-informed neural networks for solving inverse problems for the Navier–Stokes equations.Advanced Modeling and Simulation in Engineering Sciences, 10(1):5, 2023

  14. [22]

    Hy- brid data-driven closure strategies for reduced order modeling.Applied Mathematics and Computation, 448:127920, 2023

    Anna Ivagnes, Giovanni Stabile, Andrea Mola, Traian Iliescu, and Gianluigi Rozza. Hy- brid data-driven closure strategies for reduced order modeling.Applied Mathematics and Computation, 448:127920, 2023

  15. [23]

    Pressure data-driven variational multiscale reduced order models.Journal of Computational Physics, 476:111904, 2023

    Anna Ivagnes, Giovanni Stabile, Andrea Mola, Traian Iliescu, and Gianluigi Rozza. Pressure data-driven variational multiscale reduced order models.Journal of Computational Physics, 476:111904, 2023

  16. [24]

    Rutzmoser, and Daniel J

    Shobhit Jain, Paolo Tiso, Johannes B. Rutzmoser, and Daniel J. Rixen. A quadratic man- ifold for model order reduction of nonlinear structural dynamics.Computers & Structures, 188:80–94, August 2017

  17. [25]

    Error analysis and estimation in the finite volume method with applications to fluid flows

    Hrvoje Jasak. Error analysis and estimation in the finite volume method with applications to fluid flows. 1996

  18. [26]

    Mionet: Learning multiple-input operators via tensor product.SIAM Journal on Scientific Computing, 44(6):A3490–A3514, 2022

    Pengzhan Jin, Shuai Meng, and Lu Lu. Mionet: Learning multiple-input operators via tensor product.SIAM Journal on Scientific Computing, 44(6):A3490–A3514, 2022

  19. [27]

    Nonlinear model order reduction via lifting transfor- mations and proper orthogonal decomposition.AIAA Journal, 57(6):2297–2307, 2019

    Boris Kramer and Karen E Willcox. Nonlinear model order reduction via lifting transfor- mations and proper orthogonal decomposition.AIAA Journal, 57(6):2297–2307, 2019

  20. [28]

    Lumley J. L. The structure of inhomogeneous turbulent flows.Atmospheric Turbulence and Radio Wave Propagation, 1967

  21. [29]

    Model reduction of dynamical systems on nonlin- ear manifolds using deep convolutional autoencoders.Journal of Computational Physics, 404:108973, 2020

    Kookjin Lee and Kevin T Carlberg. Model reduction of dynamical systems on nonlin- ear manifolds using deep convolutional autoencoders.Journal of Computational Physics, 404:108973, 2020

  22. [30]

    Learn- ing nonlinear operators via deeponet based on the universal approximation theorem of op- erators.Nature Machine Intelligence, 3(3):218–229, Mar 2021

    Lu Lu, Pengzhan Jin, Guofei Pang, Zhongqiang Zhang, and George Em Karniadakis. Learn- ing nonlinear operators via deeponet based on the universal approximation theorem of op- erators.Nature Machine Intelligence, 3(3):218–229, Mar 2021

  23. [31]

    Springer, 2016

    Fadl Moukalled, Luca Mangani, Marwan Darwish, F Moukalled, L Mangani, and M Dar- wish.The finite volume method. Springer, 2016

  24. [32]

    A gaussian process regression approach within a data-driven pod framework for engineering problems in fluid dynamics.Mathe- matics in Engineering, 4(3):1–16, 2022

    Giulio Ortali, Nicola Demo, Gianluigi Rozza, et al. A gaussian process regression approach within a data-driven pod framework for engineering problems in fluid dynamics.Mathe- matics in Engineering, 4(3):1–16, 2022

  25. [33]

    Karl Pearson. Liii. on lines and planes of closest fit to systems of points in space.The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 2(11):559– 572, 1901

  26. [34]

    Data-driven operator inference for nonintrusive projection-based model reduction.Computer Methods in Applied Mechanics and Engineer- ing, 306:196–215, July 2016

    Benjamin Peherstorfer and Karen Willcox. Data-driven operator inference for nonintrusive projection-based model reduction.Computer Methods in Applied Mechanics and Engineer- ing, 306:196–215, July 2016

  27. [35]

    A graph convolutional autoencoder ap- proach to model order reduction for parametrized pdes.Journal of Computational Physics, 501:112762, 2024

    Federico Pichi, Beatriz Moya, and Jan S Hesthaven. A graph convolutional autoencoder ap- proach to model order reduction for parametrized pdes.Journal of Computational Physics, 501:112762, 2024

  28. [36]

    Large-eddy simulation: achievements and challenges.Progress in aerospace sciences, 35(4):335–362, 1999

    Ugo Piomelli. Large-eddy simulation: achievements and challenges.Progress in aerospace sciences, 35(4):335–362, 1999

  29. [37]

    Pope.Turbulent Flows

    Stephen B. Pope.Turbulent Flows. Cambridge University Press, 2000

  30. [38]

    Springer, 2015

    Alfio Quarteroni, Andrea Manzoni, and Federico Negri.Reduced basis methods for partial differential equations: an introduction, volume 92. Springer, 2015. 19

  31. [39]

    Springer, 2014

    Alfio Quarteroni and Gianluigi Rozza.Reduced order methods for modeling and computa- tional reduction, volume 9. Springer, 2014

  32. [40]

    Osborne Reynolds. IV. On the dynamical theory of incompressible viscous fluids and the de- termination of the criterion.Philosophical transactions of the Royal Society of London.(a.), (186):123–164, 1895

  33. [41]

    Non-linear manifold reduced- order models with convolutional autoencoders and reduced over-collocation method.Journal of Scientific Computing, 94(3):74, Feb 2023

    Francesco Romor, Giovanni Stabile, and Gianluigi Rozza. Non-linear manifold reduced- order models with convolutional autoencoders and reduced over-collocation method.Journal of Scientific Computing, 94(3):74, Feb 2023

  34. [42]

    SIAM, 2022

    Gianluigi Rozza, Giovanni Stabile, and Francesco Ballarin.Advanced reduced order methods and applications in computational fluid dynamics. SIAM, 2022

  35. [43]

    Springer Science & Business Media, 2006

    Pierre Sagaut.Large eddy simulation for incompressible flows: an introduction. Springer Science & Business Media, 2006

  36. [44]

    Driven cavity flows by efficient numerical techniques

    Rob Schreiber and Herbert B Keller. Driven cavity flows by efficient numerical techniques. Journal of Computational Physics, 49(2):310–333, 1983

  37. [45]

    Greedy construction of quadratic mani- folds for nonlinear dimensionality reduction and nonlinear model reduction.arXiv preprint arXiv:2403.06732, 2024

    Paul Schwerdtner and Benjamin Peherstorfer. Greedy construction of quadratic mani- folds for nonlinear dimensionality reduction and nonlinear model reduction.arXiv preprint arXiv:2403.06732, 2024

  38. [46]

    Proper orthogonal decom- position closure models for turbulent flows: a numerical comparison.Computer Methods in Applied Mechanics and Engineering, 237:10–26, 2012

    Zhu Wang, Imran Akhtar, Jeff Borggaard, and Traian Iliescu. Proper orthogonal decom- position closure models for turbulent flows: a numerical comparison.Computer Methods in Applied Mechanics and Engineering, 237:10–26, 2012

  39. [47]

    Data-driven filtered reduced order modeling of fluid flows.SIAM Journal on Scientific Computing, 40(3):B834–B857, 2018

    Xuping Xie, Muhammad Mohebujjaman, Leo G Rebholz, and Traian Iliescu. Data-driven filtered reduced order modeling of fluid flows.SIAM Journal on Scientific Computing, 40(3):B834–B857, 2018. 20

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.