Pith. sign in

REVIEW 3 major objections 4 minor 68 references

Towards Physics Constrained Deep Learning Based Turbulence Model Uncertainty Quantification

T0 review · 3 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A CNN learns the local gap between RANS and DNS turbulent kinetic energy; the gap is proposed to set the EPM perturbation magnitude for tighter uncertainty bounds.

desk verdict A TKE-correction CNN is real but the advertised UQ link to EPM is missing, so the paper overclaims and should be reframed or rejected. read the letter →

arxiv 2509.03833 v1 pith:ABR4RU7L submitted 2025-09-04 physics.flu-dyn

Minghan Chu , Weicheng Qian This is my paper · ORCID
classification physics.flu-dyn
keywords turbulencemodeluncertaintyeigenspaceperturbationmethodconvolutionalneuralnetworkReynolds-averagedNavier-StokesturbulentkineticenergySD7003airfoilperiodichillsmodel-form
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 proposes a way to tighten the Eigenspace Perturbation Method (EPM), the standard physics-based tool for estimating how wrong a RANS turbulence model's Reynolds-stress prediction might be. EPM knows how to perturb the stress eigenspace but not how much to perturb, so its uncertainty bounds are often far wider than needed. The authors train a one-dimensional convolutional neural network on paired RANS and DNS data to predict the local gap between modeled and true turbulent kinetic energy, intending that gap to set the local perturbation magnitude. On a stalled SD7003 airfoil and on periodic-hill flows, the CNN-corrected kinetic-energy profiles are much closer to DNS than the raw RANS profiles, with L1 errors often one to two orders of magnitude smaller. The improvement weakens for hill cases with stronger adverse pressure gradients. The paper demonstrates the correction signal but does not actually run EPM with CNN-set perturbations or compute an uncertainty bound; the UQ claim is an intended consequence, not a demonstrated result.

What carries the argument

The central object is the 1D convolutional neural network trained on paired RANS and DNS kinetic-energy profiles, combined with the EPM barycentric-triangle constraint. The CNN acts as a discrepancy function: given a RANS k profile at an x-location, it outputs a corrected k profile, and the local deviation from RANS is meant to set the EPM perturbation magnitude, the scalar that moves barycentric coordinates from the model state toward a vertex of the barycentric triangle of realizable turbulence anisotropy. The barycentric map supplies the realizability constraint that defines which perturbed stress states are physically permissible.

What would settle it

Run the proposed pipeline end to end on the SD7003 case: map the CNN-corrected k field to a spatially varying EPM perturbation magnitude, propagate it through the RANS solver, and compare the resulting uncertainty interval with the DNS profiles. If the CNN-based interval fails to contain the DNS mean-flow or pressure data in the separation bubble or at reattachment, or if no explicit mapping from corrected k to the perturbation magnitude can be written, the central claim is falsified.

Watch

Extended reading notes

Core claim

The paper's central contribution is the correction-signal idea: a one-dimensional CNN maps a wall-normal profile of RANS turbulent kinetic energy at a given streamwise location to the corresponding DNS profile, and the learned discrepancy is proposed as the spatial modulation function for EPM's barycentric perturbation magnitude. The authors show that a small CNN, trained on less than twenty percent of the streamwise locations, generalizes well enough to bring predicted k+ profiles close to DNS in separated flows over an SD7003 airfoil and periodic hills. In the periodic-hill benchmark, a network trained on one hill geometry transfers to other geometries with similar or milder pressure gradi

Load-bearing premise

The paper assumes that the CNN's local correction from modeled to true turbulent kinetic energy is a sufficient signal for setting how much to perturb the turbulence model's stress state, but it never writes that link or runs an uncertainty quantification to test it.

Editorial extensions

If this is right

  • EPM perturbation magnitude can become a spatially varying field, set by the CNN's local discrepancy, instead of a single user-chosen constant.
  • The CNN-corrected kinetic energy can serve as a discrepancy marker for future uncertainty quantification, potentially shrinking EPM bounds that are currently too generous for design.
  • Sparse training data over streamwise positions suffices to reproduce DNS-like k profiles across flow cases with similar separated-flow physics, suggesting low-cost data collection for new geometries.
  • The approach inherits non-local spatial information through convolutional kernels, addressing a known limitation of single-point eddy-viscosity closures.
  • For periodic hills with much stronger adverse pressure gradients, the correction degrades, so the method needs more training coverage or additional physics constraints in that regime.

Reading between the lines

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

  • The paper's stated goal is calibrated uncertainty bounds, but its evidence is about kinetic-energy profile accuracy; a natural next test is to build an explicit mapping from the CNN correction to the EPM perturbation magnitude and measure how often the resulting interval contains DNS mean-flow quantities.
  • Correcting the trace of the Reynolds stress does not automatically correct the anisotropy eigenvalues that EPM actually perturbs; a reliable UQ link likely needs the full Reynolds-stress discrepancy rather than only the kinetic energy.
  • One could train the network directly on a UQ objective, such as interval coverage or tightness of EPM bounds, to test whether kinetic-energy discrepancy is the right supervisory signal.
  • The extrapolation failure at high hill steepness suggests an active-learning extension: add new RANS/DNS pairs only in regimes where the discrepancy network is uncertain.
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 / 4 minor

Summary. The manuscript proposes a physics-constrained deep-learning framework for RANS model-form uncertainty quantification. The authors train a one-dimensional CNN to map wall-normal RANS turbulent kinetic energy profiles (kRANS) to DNS profiles (kDNS), and claim that the learned discrepancy can modulate the perturbation magnitude ΔB in the Eigenspace Perturbation Method (EPM), yielding spatially varying, calibrated uncertainty bounds. The paper reports L1-error improvements for k+ profiles on an SD7003 airfoil and on periodic hills (including α-sweep generalization), but it does not implement or evaluate any EPM uncertainty propagation. The central advertised UQ result is absent from the manuscript.

Significance. A method that learns where and how much to perturb RANS Reynolds stresses would be a valuable contribution, since EPM currently has no principled way to set the perturbation magnitude. The paper has useful components: paired RANS/DNS datasets, a simple transparent CNN architecture, held-out x-location and α generalization tests, and quantitative L1 comparisons against DNS. However, the paper never connects the CNN output to the EPM perturbation parameter, and no uncertainty bounds or calibration metrics are produced. If the work were reframed as a data-driven k-correction study, it would be a modest incremental contribution; as submitted, the abstract and conclusions claim a UQ capability that is not demonstrated.

major comments (3)
  1. [§5, Abstract] The abstract and Section 5 state that CNN predictions modulate the EPM perturbation magnitude ΔB, leading to physics-constrained UQ. This is not implemented or evaluated anywhere. Section 2.1 defines the perturbed barycentric coordinates via x* = x + ΔB(xt − x) and Eq. (2), but Section 2.2 and all of Section 4 address only the CNN’s scalar correction of k (kRANS → kDNS). No equation maps the CNN output to ΔB(x), no EPM-perturbed simulation is performed, and no uncertainty bounds or calibration metrics are reported. The claim that the approach was tested is thus unsupported; Section 5 itself concedes this by saying the results “pave the way” for future discrepancy marker functions.
  2. [§2.1–2.2] Even if a mapping were added, the proposed signal is of the wrong type. EPM perturbs the eigenvalue tensor Λ in the barycentric map, controlling the shape/anisotropy of the Reynolds stress tensor in Eq. (2). The CNN, however, learns a scalar correction to turbulent kinetic energy k, which enters Eq. (2) only through the multiplicative factor 2ρk. A k-only correction carries no directional or eigenvalue information, so it cannot determine how much the eigenvalues should be perturbed in barycentric space. This is a conceptual mismatch, not an omitted implementation detail, and it casts doubt on whether the advertised physics-constrained UQ framework can be built from the trained CNN as described.
  3. [§4, Table 2] The only quantitative evidence offered is L1(k+) at hand-picked x-locations, with no error bars, confidence intervals, or significance tests. Moreover, the text for the higher-α cases 6 and 7 reports L1(pred) values 1–2 orders of magnitude larger than L1(rans) at several locations (e.g., x/h = 2.076, 5.455, and 6.98 in Figure 9), while Table 2 claims generalization only for lower α. This contradicts the broad “efficacy” claim and prevents any quantitative statement about the reliability of the CNN corrections outside the training configuration.
minor comments (4)
  1. [General] There are frequent typos: “annd”, “model all sales of motion”, “effeccts”, “the the”, and inconsistent notation for the perturbation magnitude (ΔB vs. ∆B).
  2. [§4, Figure 5] The text lists evaluation locations x/c = 0.17, 0.25, 0.31, and 0.44, but Figure 5 labels the third panel x/c = 0.32. Please align.
  3. [§2.2/§3.1] The data split is described inconsistently: Section 2.2 says training used “less than 20% of the locations along the x-axis”, while Section 3.1 says “80% of the data points from all x-locations” go to training and the remaining 20% to testing. Clarify the actual holdout fraction in terms of x-locations.
  4. [Title/§1] The phrase “physics constrained” is not currently justified: no physics constraint is imposed on the CNN during training; the EPM is only mentioned as a future use. Either add a constraint to the training procedure or soften the terminology.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the CNN correction is a supervised regression against independent DNS, and the advertised EPM–ΔB link is missing rather than circular.

full rationale

The paper's core empirical step—training a 1D CNN to map kRANS to kDNS and testing on held-out x/c and x/h locations (Figs. 5–10)—is a standard supervised regression with independent DNS targets. It does not reduce to its inputs: the CNN is not fitted to the quantity it later 'predicts' in a way that makes the test result tautological. The EPM equations in Sec. 2.1 are imported from prior external work (ref. 32), and ΔB is defined but never fitted in this paper. The only self-citations (refs. 48 and 62, both by the first author) concern the SD7003 dataset and prior RANS-uncertainty studies; they are used for data provenance and motivation, not as the load-bearing justification for the new CNN result. The genuine weakness is that the paper never connects the CNN's k-correction to the EPM perturbation magnitude ΔB, and no UQ bound is computed; the abstract's claim that the approach is 'tested' for UQ efficacy is unsupported, and Sec. 5 itself concedes the work only 'paves the way.' That is an omitted deliverable, not a circular derivation. Accordingly, no step in the claimed chain is equivalent to its own input by construction.

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

The paper contributes a learned mapping from RANS to DNS kinetic energy with 86 fitted parameters and several hand choices. It borrows the EPM framework and DNS data from prior work and postulates, without evidence, that the learned correction can serve as the EPM perturbation amplitude.

free parameters (4)
  • CNN weights and biases = 86 parameters learned from paired RANS-DNS data
    The mapping from RANS k to DNS k is determined entirely by these learned weights.
  • CNN hyperparameters = kernel size 3, stride 1, valid padding, 2 FC layers, max pooling, Adam, lr 0.001, batch size 10, 800 epochs, window size
    Selected by manual exploration on a validation set, not by an external principle.
  • Evaluation x-locations = 4 per flow case (e.g., x/c = 0.17, 0.25, 0.31, 0.44; x/h ≈ 2, 3.5, 4-5, 6.8)
    Hand-picked to lie in the separation bubble and reattachment region where RANS errors are largest; this selection affects the reported L1 improvements.
  • Smoothing window = 6-point moving average applied to k+ profiles before computing L1
    Smoothing was applied to both DNS and predicted profiles; without it the reported L1 values could differ.
assumptions (4)
  • domain assumption Eigenspace Perturbation Method (EPM) as formulated by Iaccarino et al. (2017) is a valid physics constraint
    Section 2.1 uses EPM's barycentric triangle and perturbation form without re-derivation.
  • domain assumption DNS data are ground truth for turbulent kinetic energy
    The paper evaluates its CNN exclusively against DNS from refs. [59,63].
  • ad hoc to paper The RANS-to-DNS correction is learnable from a 1D local window of kRANS along the wall-normal direction
    Section 2.2 chooses a 1D CNN with window size 11 to map kRANS to kDNS; no evidence is given that local 1D windows capture the relevant non-local physics.
  • ad hoc to paper The CNN's k correction can act as the EPM modulation function
    Section 5 asserts this without any defined equation or test.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Towards Physics Constrained Deep Learning Based Turbulence Model Uncertainty Quantification." pith.science (2026). https://pith.science/paper/ABR4RU7L

@misc{pith2026250903833,
  author       = {Pith},
  title        = {Pith review of: Towards Physics Constrained Deep Learning Based Turbulence Model Uncertainty Quantification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ABR4RU7L}},
  note         = {Machine review of arXiv:2509.03833}
}
read the original abstract

Turbulence Models represent the workhorse for simulations used in engineering design and analysis. Despite their low computational cost and robustness, these models suffer from substantial predictive uncertainty, most of which is epistemic. At present, the Eigenspace Perturbation Method (EPM) is the only approach to estimate these turbulence model uncertainties, using physics based perturbation to the predicted Reynolds stresses. While the EPM address the question of how to perturb the Reynolds stresses for uncertainty estimation, it does not address how much to perturb. This shortcoming leads to very generous uncertainty bounds that result in sub-optimal designs. In this investigation, we use Convolutional Neural Networks (CNN) to predict the discrepancy between predicted and actual turbulent flows. These can be utilized to modulate the degree of the perturbations in the EPM leading to a Physics Constrained Deep Learning approach for Reynolds Averaged Navier Stokes model uncertainty quantification. We test this approach on turbulent flows over aero-foils and periodic hills to show the efficacy of our approach.

Figures

Figures reproduced from arXiv: 2509.03833 by the authors.

Figure 1
Figure 1. With freestream (U∞) coming into contact with the leading edge at 8 ◦ AoA, a three-dimensional representation of the computational domain is shown. 2.2 Convolutional Neural Network model details In this study, a one-dimensional convolutional neural network is used. The architecture of the model is selected using manual testing, and has four layers and 86 parameters in total, all of which are arranged to train a pred… view at source ↗
Figure 2
Figure 2. An illustration of the 2D periodic hills utilized in this study, with emphasis on the reference [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The methodology used in this study. The training paradigm is represented by the blue path, and [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: The flow over the SD7003 airfoil used in this study is shown, with [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Results for the dataset SD7003. CNN forecasts for normalized turbulence kinetic energy based on DNS. The first row shows the CNN-predicted DNS (pred) and ground truth (dns), which have been smoothed using a moving average with a six-step window size. Second row: Compar…
Figure 6
Figure 6. Figure 6: Tests for periodic hills in the Voet dataset. CNN prediction for normalized turbulence kinetic energy (k +) based on case 2. First row: A moving average with a six-step window size was used to smooth the corrected DNS for Case 1 (pred) against the ground truth for Case…
Figure 7
Figure 7. Figure 7: Observations for the case 2-based CNN prediction for normalized turbulence kinetic energy ( [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: Results for the case2-based CNN prediction for normalized turbulence kinetic energy ( [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Outcomes for the case2-based CNN prediction for normalized turbulence kinetic energy ( [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Results for the case2-based CNN prediction for normalized turbulence kinetic energy ( [PITH_FULL_IMAGE:figures/full_fig_p012_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 67 canonical work pages

  1. [1]

    Direct numerical simulation: a tool in turbulence research

    Parviz Moin and Krishnan Mahesh. Direct numerical simulation: a tool in turbulence research. Annual review of fluid mechanics, 30(1):539–578, 1998

    work page 1998

  2. [2]

    Direct numerical simulation of turbulent channel flow up to reτ= 590

    Robert D Moser, John Kim, and Nagi N Mansour. Direct numerical simulation of turbulent channel flow up to reτ= 590. Physics of fluids, 11(4):943–945, 1999

    work page 1999

  3. [3]

    Direct numerical simulation of turbulent channel flow up to

    Myoungkyu Lee and Robert D Moser. Direct numerical simulation of turbulent channel flow up to. Journal of fluid mechanics, 774:395–415, 2015

    work page 2015

  4. [4]

    New trends in large-eddy simulations of turbulence

    Marcel Lesieur and Olivier Metais. New trends in large-eddy simulations of turbulence. Annual review of fluid mechanics, 28(1):45–82, 1996

    work page 1996

  5. [5]

    Large-eddy simulations of turbulence

    Marcel Lesieur, Olivier Métais, and Pierre Comte. Large-eddy simulations of turbulence . Cambridge university press, 2005

    work page 2005

  6. [6]

    Analytical methods for the development of reynolds-stress closures in turbulence

    Charles G Speziale. Analytical methods for the development of reynolds-stress closures in turbulence. Annual review of fluid mechanics, 23(1):107–157, 1991

    work page 1991

  7. [7]

    Progress in the development of a reynolds-stress turbulence closure

    Brian Edward Launder, G Jr Reece, and W Rodi. Progress in the development of a reynolds-stress turbulence closure. Journal of fluid mechanics, 68(3):537–566, 1975

    work page 1975

  8. [8]

    Toward approximating non-local dynamics in single-point pressure–strain correlation closures

    Aashwin A Mishra and Sharath S Girimaji. Toward approximating non-local dynamics in single-point pressure–strain correlation closures. Journal of Fluid Mechanics, 811:168–188, 2017

    work page 2017

Show all 68 references
  1. [9]

    Modelling of rapid pressure—strain in reynolds-stress closures

    Arne V Johansson and Magnus Hallbäck. Modelling of rapid pressure—strain in reynolds-stress closures. Journal of Fluid Mechanics, 269:143–168, 1994

    1994

  2. [10]

    Modelling the pressure–strain correlation of turbulence: an invariant dynamical systems approach

    Charles G Speziale, Sutanu Sarkar, and Thomas B Gatski. Modelling the pressure–strain correlation of turbulence: an invariant dynamical systems approach. Journal of fluid mechanics, 227:245–272, 1991

    1991

  3. [11]

    Development and application of a cubic eddy-viscosity model of turbulence

    TJ Craft, BE Launder, and K Suga. Development and application of a cubic eddy-viscosity model of turbulence. International Journal of Heat and Fluid Flow, 17(2):108–115, 1996

    1996

  4. [12]

    Linear and nonlinear eddy viscosity models

    TB Gatski and CL Rumsey. Linear and nonlinear eddy viscosity models. Closure strategies for turbulent and transitional flows, pages 9–46, 2002

    2002

  5. [13]

    A new k- ϵ eddy viscosity model for high reynolds number turbulent flows

    Tsan-Hsing Shih, William W Liou, Aamir Shabbir, Zhigang Yang, and Jiang Zhu. A new k- ϵ eddy viscosity model for high reynolds number turbulent flows. Computers & fluids, 24(3):227–238, 1995

    1995

  6. [14]

    Eddy viscosity in two and three dimensions

    Robert H Kraichnan. Eddy viscosity in two and three dimensions. Journal of Atmospheric Sciences , 33(8):1521–1536, 1976

    1976

  7. [15]

    The numerical computation of turbulent flows

    Brian Edward Launder and Dudley Brian Spalding. The numerical computation of turbulent flows. In Numerical prediction of flow, heat transfer, turbulence and combustion, pages 96–116. Elsevier, 1983

    1983

  8. [16]

    Formulation of the kw turbulence model revisited

    David C Wilcox. Formulation of the kw turbulence model revisited. AIAA journal, 46(11):2823–2838, 2008

    2008

  9. [17]

    Uncertainty quantification: theory, implementation, and applications, volume 12

    Ralph C Smith. Uncertainty quantification: theory, implementation, and applications, volume 12. Siam, 2013

    2013

  10. [18]

    Turbulence modeling in the age of data.Annual Review of Fluid Mechanics, 51:357–377, 2019

    Karthik Duraisamy, Gianluca Iaccarino, and Heng Xiao. Turbulence modeling in the age of data.Annual Review of Fluid Mechanics, 51:357–377, 2019

    2019

  11. [19]

    Status, emerging ideas and future directions of turbulence modeling research in aeronautics

    Karthik Duraisamy, Philippe R Spalart, and Christopher L Rumsey. Status, emerging ideas and future directions of turbulence modeling research in aeronautics. Technical report, 2017

    2017

  12. [20]

    Bayesian uncertainty quantification applied to rans turbulence models

    Todd A Oliver and Robert D Moser. Bayesian uncertainty quantification applied to rans turbulence models. In Journal of Physics: Conference Series, volume 318, page 042032. IOP Publishing, 2011

    2011

  13. [21]

    Sensitivity of flow evolution on turbulence structure

    Aashwin A Mishra, Gianluca Iaccarino, and Karthik Duraisamy. Sensitivity of flow evolution on turbulence structure. Physical Review Fluids, 1(5):052402, 2016

    2016

  14. [22]

    Scalable environment for quantification of uncertainty and optimization in industrial applications (sequoia)

    Juan J Alonso, Michael S Eldred, Paul Constantine, Karthikeyan Duraisamy, Charbel Farhat, Gianluca Iaccarino, and John Jakeman. Scalable environment for quantification of uncertainty and optimization in industrial applications (sequoia). In 19th AIAA Non-Deterministic Approach...

    2017

  15. [23]

    Approach for uncertainty of turbulence modeling based on data assimilation technique

    Hiroshi Kato and Shigeru Obayashi. Approach for uncertainty of turbulence modeling based on data assimilation technique. Computers & Fluids, 85:2–7, 2013

    2013

  16. [24]

    Quantification of structural uncertainties in the k-w turbulence model

    Eric Dow and Qiqi Wang. Quantification of structural uncertainties in the k-w turbulence model. In 52nd AIAA/ASME/ASCE/AHS/ASC Structures, structural dynamics and materials conference 19th AIAA/ASME/AHS adaptive structures conference 13t, page 1762, 2011

    2011

  17. [25]

    Modeling imprecision and uncertainty in preliminary engineer- ing design

    Kristin L Wood and Erik K Antonsson. Modeling imprecision and uncertainty in preliminary engineer- ing design. Mechanism and Machine Theory, 25(3):305–324, 1990

    1990

  18. [26]

    Methodology for managing the effect of uncertainty in simulation-based design

    Xiaoping Du and Wei Chen. Methodology for managing the effect of uncertainty in simulation-based design. AIAA journal, 38(8):1471–1478, 2000

    2000

  19. [27]

    Robust multiattribute decision making under risk and uncer- tainty in engineering design

    Ashwin P Gurnani and Kemper Lewis. Robust multiattribute decision making under risk and uncer- tainty in engineering design. Engineering Optimization, 37(8):813–830, 2005

    2005

  20. [28]

    Comparative analysis of uncertainty propagation methods for robust engineering design

    Padulo Mattia, Campobasso Michele Sergio, and Guenov Marin Dimitrov. Comparative analysis of uncertainty propagation methods for robust engineering design. Guidelines for a Decision Support Method Adapted to NPD Processes, 2007

    2007

  21. [29]

    Impact of uncertainty quan- tification on design: an engine optimisation case study

    Michael Kokkolaras, Zissimos P Mourelatos, and Panos Y Papalambros. Impact of uncertainty quan- tification on design: an engine optimisation case study. International Journal of Reliability and Safety , 1(1-2):225–237, 2006

    2006

  22. [30]

    Review of uncertainty-based multidisciplinary design optimization methods for aerospace vehicles

    Wen Yao, Xiaoqian Chen, Wencai Luo, Michel Van Tooren, and Jian Guo. Review of uncertainty-based multidisciplinary design optimization methods for aerospace vehicles. Progress in Aerospace Sciences, 47(6):450–479, 2011

    2011

  23. [31]

    Aerospace applications of optimization under uncer- tainty

    Sharon L Padula, Clyde R Gumbert, and Wu Li. Aerospace applications of optimization under uncer- tainty. Optimization and engineering, 7:317–328, 2006

    2006

  24. [32]

    Eigenspace perturbations for uncer- tainty estimation of single-point turbulence closures

    Gianluca Iaccarino, Aashwin Ananda Mishra, and Saman Ghili. Eigenspace perturbations for uncer- tainty estimation of single-point turbulence closures. Physical Review Fluids, 2(2):024605, 2017

    2017

  25. [33]

    Uncertainty estimation module for turbulence model predictions in su2

    Aashwin Ananda Mishra, Jayant Mukhopadhaya, Gianluca Iaccarino, and Juan Alonso. Uncertainty estimation module for turbulence model predictions in su2. AIAA Journal, 57(3):1066–1077, 2019

    2019

  26. [34]

    Rans predictions for high-speed flows using enveloping models

    AA Mishra and G Iaccarino. Rans predictions for high-speed flows using enveloping models. arXiv preprint arXiv:1704.01699, 2017

    2017 arXiv

  27. [35]

    Estimating uncertainty in homogeneous turbulence evolution due to coarse-graining

    Aashwin Ananda Mishra, Karthik Duraisamy, and Gianluca Iaccarino. Estimating uncertainty in homogeneous turbulence evolution due to coarse-graining. Physics of Fluids, 31(2):025106, 2019

    2019

  28. [36]

    Uncertainty estimation for reynolds-averaged navier– stokes predictions of high-speed aircraft nozzle jets

    Aashwin Ananda Mishra and Gianluca Iaccarino. Uncertainty estimation for reynolds-averaged navier– stokes predictions of high-speed aircraft nozzle jets. AIAA Journal, 55(11):3999–4004, 2017

    2017

  29. [37]

    Eigenvector perturbation methodology for uncertainty quantification of turbulence models

    Roney L Thompson, Aashwin Ananda Mishra, Gianluca Iaccarino, Wouter Edeling, and Luiz Sampaio. Eigenvector perturbation methodology for uncertainty quantification of turbulence models. Physical Review Fluids, 4(4):044603, 2019

    2019

  30. [38]

    Robust shape optimization under model uncertainty of an aircraft wing using proper orthogonal decomposition and inductive design exploration method

    Gorkem Demir, Recep M Gorguluarslan, and Selin Aradag. Robust shape optimization under model uncertainty of an aircraft wing using proper orthogonal decomposition and inductive design exploration method. Structural and Multidisciplinary Optimization, 66(4):93, 2023

    2023

  31. [39]

    Optimization under turbulence model uncertainty for aerospace design

    Laurence W Cook, AA Mishra, JP Jarrett, KE Willcox, and G Iaccarino. Optimization under turbulence model uncertainty for aerospace design. Physics of Fluids, 31(10):105111, 2019

    2019

  32. [40]

    Design exploration and optimization under uncertainty

    Aashwin Ananda Mishra, Jayant Mukhopadhaya, Juan Alonso, and Gianluca Iaccarino. Design exploration and optimization under uncertainty. Physics of Fluids, 32(8):085106, 2020

    2020

  33. [41]

    Uncertainties quantification in the prediction of the aeroelastic response of the pazy wing tunnel model

    Marcello Righi. Uncertainties quantification in the prediction of the aeroelastic response of the pazy wing tunnel model. In AIAA SCITECH 2023 Forum, page 0761, 2023

    2023

  34. [42]

    Adjoint design optimization under the uncertainty quantification of reynolds-averaged navier-stokes turbulence model

    Anna Li, Tongsheng Wang, Jianan Chen, Zhu Huang, and Guang Xi. Adjoint design optimization under the uncertainty quantification of reynolds-averaged navier-stokes turbulence model. AIAA Journal, 62(7):2589–2600, 2024

    2024

  35. [43]

    Application of uncertainty quantification of reynolds-averaged navier–stokes models in hypersonic flow

    Anna Li, Tongsheng Wang, Jianan Chen, Guang Xi, and Zhu Huang. Application of uncertainty quantification of reynolds-averaged navier–stokes models in hypersonic flow. AIAA Journal, pages 1–9, 2025. 14 A PREPRINT - S EPTEMBER 5, 2025

    2025

  36. [44]

    Epistemic uncertainty quantification for reynolds-averaged navier-stokes modeling of separated flows over streamlined surfaces

    C Gorlé, S Zeoli, M Emory, J Larsson, and G Iaccarino. Epistemic uncertainty quantification for reynolds-averaged navier-stokes modeling of separated flows over streamlined surfaces. Physics of Fluids, 31(3):035101, 2019

    2019

  37. [45]

    A nonuniform perturbation to quantify rans model uncertainties

    Z Huang, A Mishra, and G Iaccarino. A nonuniform perturbation to quantify rans model uncertainties. Center for Turbulence Research Annual Research Briefs, Stanford Univ., Stanford, CA, pages 223–232, 2020

    2020

  38. [46]

    Multi- fidelity modeling of probabilistic aerodynamic databases for use in aerospace engineering

    Jayant Mukhopadhaya, Brian T Whitehead, John F Quindlen, Juan J Alonso, and Andrew W Cary. Multi- fidelity modeling of probabilistic aerodynamic databases for use in aerospace engineering. International Journal for Uncertainty Quantification, 10(5), 2020

    2020

  39. [47]

    A toolset for creation of multi-fidelity probabilistic aerodynamic databases

    Nikhil Nigam, Seyedmohammad Mohseni, Jonathan Valverde, Sergey Voronin, Jayant Mukhopadhaya, and Juan J Alonso. A toolset for creation of multi-fidelity probabilistic aerodynamic databases. In AIAA Scitech 2021 Forum, page 0466, 2021

    2021

  40. [48]

    Model-form uncertainty quantification of reynolds- averaged navier–stokes modeling of flows over a sd7003 airfoil

    Minghan Chu, Xiaohua Wu, and David E Rival. Model-form uncertainty quantification of reynolds- averaged navier–stokes modeling of flows over a sd7003 airfoil. Physics of Fluids, 34(11):117105, 2022

    2022

  41. [49]

    Machine learning for fluid mechanics

    Steven L Brunton, Bernd R Noack, and Petros Koumoutsakos. Machine learning for fluid mechanics. Annual Review of Fluid Mechanics, 52:477–508, 2020

    2020

  42. [50]

    Data-assisted combustion simulations with dynamic submodel assignment using random forests

    Wai Tong Chung, Aashwin Ananda Mishra, Nikolaos Perakis, and Matthias Ihme. Data-assisted combustion simulations with dynamic submodel assignment using random forests. Combustion and Flame, 227:172–185, 2021

    2021

  43. [51]

    Interpretable data-driven methods for subgrid-scale closure in les for transcritical lox/gch4 combustion

    Wai Tong Chung, Aashwin Ananda Mishra, and Matthias Ihme. Interpretable data-driven methods for subgrid-scale closure in les for transcritical lox/gch4 combustion. Combustion and Flame, 239:111758, 2022

    2022

  44. [52]

    Perspectives on machine learning-augmented reynolds-averaged and large eddy simulation models of turbulence

    Karthik Duraisamy. Perspectives on machine learning-augmented reynolds-averaged and large eddy simulation models of turbulence. Physical Review Fluids, 6(5):050504, 2021

    2021

  45. [53]

    Machine learning methods for data-driven turbulence modeling

    Ze Jia Zhang and Karthikeyan Duraisamy. Machine learning methods for data-driven turbulence modeling. In 22nd AIAA computational fluid dynamics conference, page 2460, 2015

    2015

  46. [54]

    Workshop report on basic research needs for scientific machine learning: Core technologies for artificial intelligence

    Nathan Baker, Frank Alexander, Timo Bremer, Aric Hagberg, Yannis Kevrekidis, Habib Najm, Manish Parashar, Abani Patra, James Sethian, Stefan Wild, et al. Workshop report on basic research needs for scientific machine learning: Core technologies for artificial intelligence. Tec...

    2019

  47. [55]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations

    Maziar Raissi, Paris Perdikaris, and George E Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational physics, 378:686–707, 2019

    2019

  48. [56]

    Physics constrained unsupervised deep learning for rapid, high resolution scanning coherent diffraction reconstruction

    Oliver Hoidn, Aashwin Ananda Mishra, and Apurva Mehta. Physics constrained unsupervised deep learning for rapid, high resolution scanning coherent diffraction reconstruction. Scientific reports, 13(1):22789, 2023

    2023

  49. [57]

    Turbulent flows, 2001

    Stephen B Pope. Turbulent flows, 2001

    2001

  50. [58]

    Presentation of anisotropy properties of turbulence, invariants versus eigenvalue approaches

    S Banerjee, R Krahl, F Durst, and Ch Zenger. Presentation of anisotropy properties of turbulence, invariants versus eigenvalue approaches. Journal of Turbulence, (8):N32, 2007

    2007

  51. [59]

    Turbulent and Non-Turbulent Interfaces in Low Mach Number Airfoil Flows

    Huiying Zhang. Turbulent and Non-Turbulent Interfaces in Low Mach Number Airfoil Flows. PhD thesis, Queen’s University (Canada), 2021

    2021

  52. [60]

    A correlation-based transition model using local variables for unstructured parallelized cfd codes

    Robin Blair Langtry. A correlation-based transition model using local variables for unstructured parallelized cfd codes. 2006

    2006

  53. [61]

    Correlation-based transition modeling for unstructured paral- lelized computational fluid dynamics codes

    Robin B Langtry and Florian R Menter. Correlation-based transition modeling for unstructured paral- lelized computational fluid dynamics codes. AIAA journal, 47(12):2894–2906, 2009

    2009

  54. [62]

    Quantification of reynolds-averaged-navier–stokes model-form uncertainty in transitional boundary layer and airfoil flows

    Minghan Chu, Xiaohua Wu, and David E Rival. Quantification of reynolds-averaged-navier–stokes model-form uncertainty in transitional boundary layer and airfoil flows. Physics of Fluids, 34(10):107101, 2022

    2022

  55. [63]

    A hybrid approach combining dns and rans simulations to quantify uncertainties in turbulence modelling

    Laurens JA Voet, Richard Ahlfeld, Audrey Gaymann, Sylvain Laizet, and Francesco Montomoli. A hybrid approach combining dns and rans simulations to quantify uncertainties in turbulence modelling. Applied Mathematical Modelling, 89:885–906, 2021

    2021

  56. [64]

    Implicit large eddy simulation of low-reynolds-number transi- tional flow past the sd7003 airfoil

    Marshall Galbraith and Miguel Visbal. Implicit large eddy simulation of low-reynolds-number transi- tional flow past the sd7003 airfoil. In 40th fluid dynamics conference and exhibit, page 4737, 2010. 15 A PREPRINT - S EPTEMBER 5, 2025

    2010

  57. [65]

    Daniel J Garmann, Miguel R Visbal, and Paul D Orkwis. Comparative study of implicit and subgrid- scale model large-eddy simulation techniques for low-reynolds number airfoil applications.International Journal for Numerical Methods in Fluids, 71(12):1546–1565, 2013

    2013

  58. [66]

    Data driven physics constrained perturbations for turbulence model uncertainty estimation

    Jan Felix Heyse, Aashwin Ananda Mishra, and Gianluca Iaccarino. Data driven physics constrained perturbations for turbulence model uncertainty estimation. In AAAI Spring Symposium: MLPS, 2021

    2021

  59. [67]

    Estimating rans model uncertainty using machine learning

    Jan Felix Heyse, Aashwin A Mishra, and Gianluca Iaccarino. Estimating rans model uncertainty using machine learning. Journal of the Global Power and Propulsion Society, 2021(May):1–14, 2021

    2021

  60. [68]

    Evaluation of physics constrained data- driven methods for turbulence model uncertainty quantification

    Marcel Matha, Karsten Kucharczyk, and Christian Morsbach. Evaluation of physics constrained data- driven methods for turbulence model uncertainty quantification. Computers & Fluids , 255:105837, 2023. 16

    2023

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.