Pith. sign in

REVIEW 5 major objections 4 minor 52 references

Robust Airfoil Design Optimization via a Bilevel Model-Based Methodology

T0 review · 5 major / 4 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read This paper proposes GLORO, a bilevel surrogate-based method that treats worst-case robust airfoil design as two nested optimization problems and reports up to 52 percent lift-to-drag gains with 217 CFD simulations.

desk verdict A promising bilevel surrogate method with impressive CFD cost, but the 7-sample BO worst-case estimate is unvalidated and the 'consistently improved' claim is contradicted by the paper's own table. read the letter →

arxiv 2607.29161 v1 pith:WS2DJM4T submitted 2026-07-31 math.OC

Alpaslan Kurt , Dilara Buk , Yusuf Yazicioglu , Figen Oztoprak , Onur Son , Gorkem Demir This is my paper · ORCID
classification math.OC MSC 90C5690C2690C47
keywords robustoptimizationexpensiveblack-boxfunctionsBayesianGaussianprocesssurrogatesderivative-freetrust-regionmethodsairfoildesignlift-to-dragratio
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

The paper tries to show that worst-case robust optimization of a computationally expensive black-box function can be made affordable by splitting it into two nested surrogate problems. In the lower level, the worst case over uncertain parameters (Mach number and angle of attack) is approximated by Bayesian optimization on a Gaussian-process model, with the GP posterior as a byproduct. In the upper level, those noisy approximate worst-case values feed local quadratic trust-region models in the 30-dimensional design space, using one random sample point to keep the models useful without a full sample budget. On the transonic airfoil test case, the method completes in 217 CFD simulations and produces a shape whose worst-case lift-to-drag ratio improves by up to 52 percent at specific conditions while keeping low variance across the uncertainty box. If true, this means robust designs can be sought with budgets comparable to a handful of deterministic optimizations, without adjoint gradients.

What carries the argument

The load-bearing mechanism is a two-level surrogate architecture. Lower level: for a fixed design, the robust constraint c_R(x)=min_p c(x;p) is solved approximately by Bayesian optimization over the parameter box; the GP regression model constructed along the way is the uncertainty-quantification output, so no separate UQ phase is needed. Upper level: a model-based derivative-free trust-region algorithm fits quadratic local models of the robust objective and thickness constraint in design space from a small, underdetermined sample set, using minimum-norm regression and one randomized point per iteration to raise rank; the trust region can shrink but never below a floor, reflecting that model

What would settle it

Run the same CFD solver on the optimized airfoil over a dense grid (e.g., 20×20) of (M, alpha) in [0.715,0.735] × [2.82,3.02] degrees and compare each point's L/D with the worst-case value that GLORO's inner Bayesian optimization reported during optimization. A grid point whose L/D is materially below the reported robust value—or below the baseline airfoil's worst-case L/D—would show that the five-iteration BO approximation missed the true worst case and that the claimed robustness is an artifact.

Watch

Extended reading notes

Core claim

Central claim: for robust problems of the form max_x min_{p in U} c(x;p) with expensive c, GLORO's pairing of Bayesian-optimized GP surrogates in parameter space with randomized model-based trust-region optimization in design space yields acceptable robust solutions with far fewer simulations than global-surrogate or multipoint alternatives. For a transonic airfoil with 30 bump-shape design variables and uncertainty sets of ±0.01 in Mach number and ±0.1 degrees in angle of attack, the paper reports 10 outer iterations and 217 RANS simulations. The optimized geometry improves the worst-case lift-to-drag ratio by up to 52 percent near the high-Mach boundary, with drag reductions up to 34 perce

Load-bearing premise

The load-bearing premise is that a Bayesian optimization run of two initial points plus five acquisition iterations gives a trustworthy estimate of the worst-case lift-to-drag ratio over the whole continuous Mach–AoA box; if those estimates are off, the optimizer is chasing phantom worst-case points and the final design is not necessarily robust.

Editorial extensions

If this is right

  • Worst-case robust design of airfoils becomes possible within a few hundred CFD runs, making robust shape optimization practical in early design cycles.
  • The cost advantage should grow as the number of uncertain parameters increases, because a multipoint or grid alternative grows exponentially while GLORO's lower level remains a low-dimensional Bayesian optimization.
  • No adjoint or gradient information is required, so the method applies to black-box simulation codes where adjoints are unavailable.
  • The naturally emerging secondary compression in the optimized shape suggests that geometric tailoring alone can produce the drag-reduction mechanism of shock control bumps.

Reading between the lines

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

  • The paper's efficiency claim rests on 2+5 Bayesian-optimization iterations per robust evaluation; a natural extension is to verify the optimized shape on a dense grid or with a second optimizer to quantify how much of the reported robustness is real versus an artifact of surrogate optimism.
  • If the Bayesian-optimization approximate worst-case values are biased, the same trust-region framework could be adapted to report confidence intervals on the robust objective, which would be valuable for certification-style design.
  • The emergent shock-control-bump-like flow feature suggests a testable design hypothesis: allowing the optimizer to generate natural secondary compressions might outperform explicit bump parameterizations, a comparison the paper does not run.
  • The bilevel GLORO scheme could transfer to other expensive black-box robust design settings with a small number of uncertain parameters, such as turbomachinery or structural design, where worst-case performance over a box is the design driver.
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

5 major / 4 minor

Summary. The paper proposes GLORO, a bilevel methodology for robust optimization with expensive black-box evaluations. The lower level approximates the worst-case value of the objective over an uncertainty set by Bayesian optimization with Gaussian-process surrogates; the upper level uses trust-region local models in the design space, with approximate robust evaluations treated as noisy. The method is applied to robust optimization of the RAE2822 airfoil under uncertainty in Mach number and angle of attack, with 30 design variables and a minimum-thickness constraint. The paper reports an optimized airfoil that improves the worst-case lift-to-drag ratio over the considered flight conditions, up to 52% at one tested condition, using 217 CFD simulations. A toy HS91 problem is also used as a preliminary test.

Significance. If the reported robust evaluations are trustworthy, the methodology is interesting and potentially very sample-efficient: 217 RANS evaluations for a 30-variable worst-case optimization is far below the costs reported for comparable robust aerodynamic shape optimization studies (e.g., Table 5.2). The integration of Bayesian optimization in the parameter space with local model-based DFO in the design space is a sensible and clearly motivated novelty. The paper is also honest about the absence of convergence analysis and about the stochastic nature of the evaluations. However, the central quantitative claim depends on the accuracy of 7-point Bayesian-optimization estimates of continuous worst-case values, and this accuracy is not validated. The paper is therefore of interest, but its main claim needs additional support before it can be accepted.

major comments (5)
  1. [Section 4, Optimization; Table 4.3] The approximate robust evaluations are computed with only two initial points plus five Bayesian-optimization iterations in the 2D parameter box, i.e., 7 CFD simulations per design. The paper does not check whether these estimates are close to the true worst-case L/D over [0.715,0.735]×[2.82°,3.02°]. No dense grid, no random sampling, and no independent optimization of the parameter-space problem is reported. The final design is evaluated on only a 3×3 grid (Table 4.3), which cannot certify the continuous worst-case claim. If the 7-point GP systematically misses a sharp transonic drag-rise corner, the upper-level DFO is optimizing against phantom values. Please add a validation study: for at least the baseline and final designs (and preferably several intermediate iterates), compare the BO estimate of the worst-case value with a dense-grid or Latin-hypercube evaluation, and report the err
  2. [Section 5, Conclusions; Table 4.3] The conclusion states that the results show 'consistently improved performance in low-performing cases, including a maximum increase of 52%.' However, Table 4.3 shows that the optimized airfoil is worse than the baseline at all three angles of attack for M=0.715 (e.g., -10.29%, -6.89%, -2.61% in L/D). A max-min design may legitimately sacrifice high-performance points to improve the worst case, but the wording 'consistently improved performance' is not supported by the presented data. Please rephrase the conclusion to refer specifically to worst-case or low-end performance, and report the minimum L/D over the tested grid for both designs.
  3. [Section 3.2, 'We do not have a merit parameter update strategy'; Algorithm 2, Step 4] The algorithm assumes a fixed, sufficiently large λ for the merit function penalty on the thickness constraint, but there is no update strategy and no verification that the final design actually satisfies h(x) ≥ h_min. Since the thickness constraint is a hard constraint in problem (2.1), the paper should report the final thickness value or constraint margin, and ideally demonstrate that the chosen λ is large enough across the optimization. Without this, the reported optimized geometry may be infeasible.
  4. [Section 3.2, Algorithm 2, Step 5; Section 5] GLORO is stochastic: it uses random sample points for rank completion and the lower-level Bayesian optimization is randomized. Yet all results are from a single run, and the paper explicitly defers sensitivity to randomness to future work. For the empirical claim that GLORO 'can find acceptable solutions,' at least a small number of replications (e.g., 3-5 runs with different random seeds) or a sensitivity study is needed. A single run cannot establish the method's effectiveness or the stability of the reported 52% improvement.
  5. [Appendix A, Table 5.1] The efficiency comparison with SLSQP multipoint solves is not like-for-like. The SLSQP runs solve deterministic fixed-parameter problems, while GLORO solves the robust problem; furthermore, SLSQP uses adjoint evaluations (which are generally more expensive than primal-only evaluations). The statement that 'within the computation budget required by GLORO the multipoint approach can complete at most 3-4 solves' is therefore not a direct efficiency measure. Please clarify the purpose of this comparison and temper the efficiency claim accordingly.
minor comments (4)
  1. [Section 4, Optimization] The text says the design variables are '38 Hicks-Henne bump function parameters' but then states n=30. Please clarify this discrepancy, e.g., by describing how the 38 parameters are reduced to 30 design variables.
  2. [Section 3.1] Typographical and language issues: 'revoked approach' should likely be 'revived approach' or 'revisited approach'; 'existance' should be 'existence'; 'straigtforward' should be 'straightforward'; 'bugdet' in Appendix A should be 'budget'.
  3. [Algorithm 2, Require] The parameter θ is listed in the Require block and used in Step 5, but its role is never explained. Please state how θ is chosen (e.g., θ=2 is common in trust-region methods) and how it affects the set of points included in the local model.
  4. [Section 4, Figure 4.5] Figure 4.5 is referenced as showing variance of the lift-to-drag ratio, but the captions and axis labels are not described in the text. Please make the figure self-explanatory or describe what exactly is plotted (e.g., box plots, scatter over the 3×3 grid).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: final designs are evaluated by independent CFD and no fitted parameter is renamed as a prediction.

full rationale

The paper's claims are methodological and empirical, not derivational: GLORO is an optimization algorithm applied to a robust airfoil design problem, and the reported improvements are direct CFD comparisons between baseline and optimized geometries. The lower-level Bayesian optimization provides only approximate worst-case evaluations used to guide the search, and the paper explicitly labels these as approximate (Algorithm 2, Steps 1 and 4: 'Compute approximate robust evaluations c~R(xi) ≈ min_{p∈Up} c(xi;p) via Bayesian optimization'). The final table (Table 4.3) compares lift, drag, and lift-to-drag ratios at a 3×3 grid of Mach and angle-of-attack values for the baseline and optimized airfoils; these are independent CFD evaluations, not outputs of the GP surrogate or of the fitted local models. No parameter is fitted to the final performance numbers and then reported as a prediction; the algorithm parameters (N0 = 10, Δ0 = 0.5×10−5, λ = 10^4, ρ1 = 0.01, ρ2 = 0.6, and 2 + 5 Bayesian-optimization iterations) are implementation choices, not calibration targets. There are no load-bearing self-citations: the references for GP, Bayesian optimization, and model-based DFO are standard external works, and no 'uniqueness theorem' or prior result by the same authors is used to force the outcome. The main validity concern—that each worst-case evaluation uses only seven samples and is never checked against a dense grid or exact minimization over the continuous uncertainty box—is a question of approximation accuracy and robustness of the empirical conclusions, not circularity. A poor approximation could make the optimized design non-robust, but that is a correctness/validation gap, not an equivalence between the inputs and outputs of the claimed derivation. Accordingly, no circular step can be exhibited under the required standard.

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

No new physical entities are introduced. The algorithm has numerous hand-set parameters (trust region, merit weight, BO budget) and strong modeling assumptions about smoothness and surrogate accuracy. The free parameters do not fit a physical constant, but they do gate the reliability of the result.

free parameters (5)
  • lambda (merit penalty weight) = 10^4
    Chosen ad hoc; paper assumes a large enough value is selected (Section 3.2). Affects balance between objective and thickness constraint in local model.
  • Trust region parameters Delta0, Delta_min, gamma, rho1, rho2 = 0.5e-5, 1e-7, 2, 0.01, 0.6
    Set by hand in Section 4. Control step size and acceptance of candidate points; not derived from problem data.
  • Initial sample count N0 = 10
    Number of initial robust evaluations in the design space; chosen ad hoc (Section 4).
  • Bayesian optimization budget per robust evaluation = 2 initial GP points + 5 BO iterations
    Only 7 CFD simulations per design to approximate the worst-case min over the uncertainty box. This heavily influences accuracy and cost; no justification beyond cost saving.
  • GP hyperparameters = not reported
    Presumably estimated via MLE inside each BO run; the kernel, lengthscales, and noise variance are not stated, so their values and influence are undisclosed.
assumptions (5)
  • domain assumption The worst-case (min over uncertainty set) formulation is the appropriate robustness criterion.
    Problem (2.1) adopts min over M and alpha of CL/CD; alternative mean-variance or percentile robustness measures are possible and the paper does not justify the choice beyond referencing common robust optimization.
  • domain assumption RANS with the Spalart-Allmaras turbulence model is sufficiently accurate for the lift and drag coefficients used in optimization.
    Validated against AGARD Case 6 (Section 4), but no mesh convergence study or quantification of CFD error is provided.
  • domain assumption The robust objective is smooth enough in the 30-dimensional design space to be locally approximated by a low-degree polynomial from severely underdetermined sample sets.
    The algorithm builds quadratic models with as few as 10 points in 30 dimensions (q=496 coefficients) and minimum-norm solutions; no well-poisedness is enforced.
  • ad hoc to paper Bayesian optimization with 5 iterations returns a useful approximation of the continuous worst-case value.
    The paper explicitly calls these 'approximate robust evaluations' but does not validate their accuracy. This is the load-bearing assumption that links the surrogate-based optimization to true robustness.
  • ad hoc to paper A fixed, sufficiently large lambda in the merit function ensures thickness constraints are satisfied without an update strategy.
    Section 3.2 states: 'We assume that a large enough value of λ > 0 is chosen.' No mechanism adapts λ.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Robust Airfoil Design Optimization via a Bilevel Model-Based Methodology." pith.science (2026). https://pith.science/paper/WS2DJM4T

@misc{pith2026260729161,
  author       = {Pith},
  title        = {Pith review of: Robust Airfoil Design Optimization via a Bilevel Model-Based Methodology},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WS2DJM4T}},
  note         = {Machine review of arXiv:2607.29161}
}
read the original abstract

We propose GLORO, a bilevel solution methodology for robust optimization involving expensive function evaluations. The methodology is carefully designed to achieve satisfactory optimization results in a computationally efficient manner. It is based on Gaussian Process surrogates in the parameter space for approximate computations of the objective / constraints of the robust optimization problem via Bayesian optimization (lower level), and local models constructed in the variable space using these approximate evaluations (upper level). Both the use of Bayesian optimization (in parameter space) and the use of local models (in variable space) are motivated by the idea of guiding the expensive function evaluations to the regions of interest for the optimization process. The methodological work is motivated by and tested on a robust airfoil design optimization problem. This application focuses on the RAE2822 airfoil, optimizing its shape to ensure a robust lift-to-drag ratio under operational uncertainties in Mach number and angle of attack.

Figures

Figures reproduced from arXiv: 2607.29161 by the authors.

Figure 3.1
Figure 3.1. Evaluation of the constraint value for the robust formulation at a candidate design [PITH_FULL_IMAGE:figures/full_fig_p008_3_1.png] view at source ↗
Figure 3.2
Figure 3.2. Value of the constraint c(x; µ) for different realizations of µ1 and µ2. The cells corre￾sponding to a constraint violation are marked in red. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_3_2.png] view at source ↗
Figure 4.1
Figure 4.1. Mesh structure around the RAE 2822 airfoil. (a) full computational domain, (b) near [PITH_FULL_IMAGE:figures/full_fig_p013_4_1.png] view at source ↗
Figures from the paper (4 more)
Figure 4.2
Figure 4.2. Figure 4.2: Pressure coefficient (Cp) distribution over the RAE2822 airfoil at M = 0.725, Re = 6.5 × 106 , and CL = 0.743. Experimental results from Case 6 of Cook et al.[14] at α = 2.92◦ (red) are compared with the present numerical results at α = 2.44◦ (blue). 13 [PITH_FULL_I…
Figure 4.3
Figure 4.3. Figure 4.3: Comparison of the original RAE2822 airfoil and the optimized airfoil shapes. [PITH_FULL_IMAGE:figures/full_fig_p015_4_3.png]
Figure 4.4
Figure 4.4. Figure 4.4: Comparison of pressure coefficient distributions between the baseline and optimized airfoils at Mach numbers [PITH_FULL_IMAGE:figures/full_fig_p017_4_4.png]
Figure 4.5
Figure 4.5. Figure 4.5: Performances of the designs for different parameter realizations. [PITH_FULL_IMAGE:figures/full_fig_p018_4_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 3 linked inside Pith

  1. [1]

    https://github.com/su2code/Tutorials/tree/master/design/ Turbulent 2D Constrained RAE2822

    Resources for the tutorial turbulent 2d constrained rae2822. https://github.com/su2code/Tutorials/tree/master/design/ Turbulent 2D Constrained RAE2822

  2. [2]

    A trust-region method for derivative-free nonlinear constrained stochastic optimization.arXiv preprint arXiv:1703.04156, 2017

    F Augustin and YM Marzouk. A trust-region method for derivative-free nonlinear constrained stochastic optimization.arXiv preprint arXiv:1703.04156, 2017

    arXiv 2017

  3. [3]

    Convergence of trust-region methods based on probabilistic models.SIAM Journal on Optimization, 24(3):1238–1264, 2014

    Afonso S Bandeira, Katya Scheinberg, and Luis Nunes Vicente. Convergence of trust-region methods based on probabilistic models.SIAM Journal on Optimization, 24(3):1238–1264, 2014

    2014

  4. [4]

    Robust optimization–methodology and applications

    Aharon Ben-Tal and Arkadi Nemirovski. Robust optimization–methodology and applications. Mathematical programming, 92:453–480, 2002

    2002

  5. [5]

    Nonconvex robust optimization for problems with constraints.INFORMS journal on computing, 22(1):44–58, 2010

    Dimitris Bertsimas, Omid Nohadani, and Kwong Meng Teo. Nonconvex robust optimization for problems with constraints.INFORMS journal on computing, 22(1):44–58, 2010

    2010

  6. [6]

    A comparative study of learning techniques for the compress- ible aerodynamics over a transonic rae2822 airfoil.Computers & Fluids, 251:105759, 2023

    Giovanni Catalani, Daniel Costero, Michael Bauerheim, Luca Zampieri, Vincent Chapin, Nico- las Gourdain, and Pierre Baqu´ e. A comparative study of learning techniques for the compress- ible aerodynamics over a transonic rae2822 airfoil.Computers & Fluids, 251:105759, 2023

    2023

  7. [7]

    Further comparisons of interactive boundary-layer and thin-layer navier-stokes procedures.Journal of aircraft, 25(10):897–903, 1988

    Kuei-Chung Chang, N Alemdaroglu, Unmeel Mehta, and Tuncer Cebeci. Further comparisons of interactive boundary-layer and thin-layer navier-stokes procedures.Journal of aircraft, 25(10):897–903, 1988

    1988

  8. [8]

    Stochastic optimization using a trust- region method and random models.Mathematical Programming, 169(2):447–487, 2018

    Ruobing Chen, Matt Menickelly, and Katya Scheinberg. Stochastic optimization using a trust- region method and random models.Mathematical Programming, 169(2):447–487, 2018

    2018

Show all 52 references
  1. [9]

    Randomized derivative-free optimization of noisy convex functions.arXiv preprint arXiv:1507.03332, 2015

    Ruobing Chen and Stefan Wild. Randomized derivative-free optimization of noisy convex functions.arXiv preprint arXiv:1507.03332, 2015

    2015 arXiv

  2. [10]

    Wind tunnel wall interference correction for transonic airfoils with data-reduced ensemble kalman filter.Physics of Fluids, 36(10), 2024

    Xin Chen, Gang Wang, and Zhengyin Ye. Wind tunnel wall interference correction for transonic airfoils with data-reduced ensemble kalman filter.Physics of Fluids, 36(10), 2024

    2024

  3. [11]

    SIAM, 2009

    Andrew R Conn, Katya Scheinberg, and Luis N Vicente.Introduction to derivative-free opti- mization. SIAM, 2009

    2009

  4. [12]

    Bilevel derivative-free optimization and its applica- tion to robust optimization.Optimization Methods and Software, 27(3):561–577, 2012

    Andrew R Conn and Lu ´ ıs Nunes Vicente. Bilevel derivative-free optimization and its applica- tion to robust optimization.Optimization Methods and Software, 27(3):561–577, 2012

    2012

  5. [13]

    Robust airfoil optimization and the importance of appropriately representing uncertainty.AIAA Journal, 55(11):3925–3939, 2017

    Laurence W Cook and Jerome P Jarrett. Robust airfoil optimization and the importance of appropriately representing uncertainty.AIAA Journal, 55(11):3925–3939, 2017

    2017

  6. [14]

    Aerofoil rae 2822-pressure distributions, and boundary layer and wake measurements

    PH Cook, MA McDonald, and MCP Firmin. Aerofoil rae 2822-pressure distributions, and boundary layer and wake measurements. experimental data base for computer program assess- ment.AGARD report ar, 138:47, 1979

    1979

  7. [15]

    Learning surrogate models for simulation-based optimization.AIChE Journal, 60(6):2211–2227, 2014

    Alison Cozad, Nikolaos V Sahinidis, and David C Miller. Learning surrogate models for simulation-based optimization.AIChE Journal, 60(6):2211–2227, 2014. 21

    2014

  8. [16]

    Robust airfoil optimization using maximum expected value and expected maximum value approaches.AIAA journal, 50(9):1905–1919, 2012

    Ana-Maria Croicu, M Yousuff Hussaini, Antony Jameson, and Goetz Klopfer. Robust airfoil optimization using maximum expected value and expected maximum value approaches.AIAA journal, 50(9):1905–1919, 2012

    1905

  9. [17]

    Sensitivity and calibration of turbulence model in the presence of epistemic uncertainties.CEAS Aeronautical Journal, 11(1):33–47, 2020

    Andrea Da Ronch, Marco Panzeri, Jernej Drofelnik, and Roberto d’Ippolito. Sensitivity and calibration of turbulence model in the presence of epistemic uncertainties.CEAS Aeronautical Journal, 11(1):33–47, 2020

    2020

  10. [18]

    Metamodel-based robust simulation- optimization: An overview.Uncertainty Management in Simulation-Optimization of Complex Systems: Algorithms and Applications, pages 27–54, 2015

    Gabriella Dellino, Jack PC Kleijnen, and Carlo Meloni. Metamodel-based robust simulation- optimization: An overview.Uncertainty Management in Simulation-Optimization of Complex Systems: Algorithms and Applications, pages 27–54, 2015

    2015

  11. [19]

    Robust aerodynamic design optimization using poly- nomial chaos.Journal of Aircraft, 46(2):635–646, 2009

    Michael Dodson and Geoffrey T Parks. Robust aerodynamic design optimization using poly- nomial chaos.Journal of Aircraft, 46(2):635–646, 2009

    2009

  12. [20]

    Su2: An open-source suite for multiphysics simulation and design.Aiaa Journal, 54(3):828–846, 2016

    Thomas D Economon, Francisco Palacios, Sean R Copeland, Trent W Lukaczyk, and Juan J Alonso. Su2: An open-source suite for multiphysics simulation and design.Aiaa Journal, 54(3):828–846, 2016

    2016

  13. [21]

    John Wiley & Sons, 2008

    Alexander Forrester, Andras Sobester, and Andy Keane.Engineering design via surrogate modelling: a practical guide. John Wiley & Sons, 2008

    2008

  14. [22]

    Assess- ment of robust optimization for design of rotorcraft airfoils in forward flight.Aerospace Science and Technology, 107:106355, 2020

    Francesca Fusi, Pietro Marco Congedo, Alberto Guardone, and Giuseppe Quaranta. Assess- ment of robust optimization for design of rotorcraft airfoils in forward flight.Aerospace Science and Technology, 107:106355, 2020

    2020

  15. [23]

    Numerical study of wind-tunnel walls effects on transonic airfoil flow.AIAA journal, 41(6):1046–1054, 2003

    Andrey Garbaruk, Mikhail Shur, Mikhail Strelets, and Philippe R Spalart. Numerical study of wind-tunnel walls effects on transonic airfoil flow.AIAA journal, 41(6):1046–1054, 2003

    2003

  16. [24]

    Cambridge University Press, 2023

    Roman Garnett.Bayesian optimization. Cambridge University Press, 2023

    2023

  17. [25]

    Constrained optimization with low-rank tensors and applications to parametric problems with pdes.SIAM Journal on Scientific Computing, 39(1):A25–A54, 2017

    Sebastian Garreis and Michael Ulbrich. Constrained optimization with low-rank tensors and applications to parametric problems with pdes.SIAM Journal on Scientific Computing, 39(1):A25–A54, 2017

    2017

  18. [26]

    A practical guide to robust opti- mization.Omega, 53:124–137, 2015

    Bram L Gorissen, Ihsan Yanikoglu, and Dick Den Hertog. A practical guide to robust opti- mization.Omega, 53:124–137, 2015

    2015

  19. [27]

    Expected decrease for derivative-free algorithms using random subspaces.Mathematics of Computation, 94(351):277–304, 2025

    Warren Hare, Lindon Roberts, and Cl´ ement Royer. Expected decrease for derivative-free algorithms using random subspaces.Mathematics of Computation, 94(351):277–304, 2025

    2025

  20. [28]

    Test examples for nonlinear programming codes.Journal of optimization theory and applications, 30(1):127–129, 1980

    Willi Hock and Klaus Schittkowski. Test examples for nonlinear programming codes.Journal of optimization theory and applications, 30(1):127–129, 1980

    1980

  21. [29]

    Numerical solution of the euler equations by finite volume methods using runge kutta time stepping schemes

    Antony Jameson, Wolfgang Schmidt, and Eli Turkel. Numerical solution of the euler equations by finite volume methods using runge kutta time stepping schemes. In14th fluid and plasma dynamics conference, page 1259, 1981

    1981

  22. [30]

    D-optimal polynomial chaos expansion for adjoint-based aerodynamic robust optimization in transonic flows.Aerospace Science and Technology, page 110659, 2025

    Xinze Ji, Tihao Yang, Yayun Shi, Yuhang Ma, and Junqiang Bai. D-optimal polynomial chaos expansion for adjoint-based aerodynamic robust optimization in transonic flows.Aerospace Science and Technology, page 110659, 2025. 22

    2025

  23. [31]

    Optimisation of adaptive shock control bumps with structural constraints.Aerospace Science and Technology, 77:332–343, 2018

    Edward Jinks, Paul Bruce, and Matthew Santer. Optimisation of adaptive shock control bumps with structural constraints.Aerospace Science and Technology, 77:332–343, 2018

    2018

  24. [32]

    Cokriging for robust design optimization.AIAA journal, 50(11):2351–2364, 2012

    Andy J Keane. Cokriging for robust design optimization.AIAA journal, 50(11):2351–2364, 2012

    2012

  25. [33]

    Robust design optimization using surrogate models

    Andy J Keane and Ivan I Voutchkov. Robust design optimization using surrogate models. Journal of Computational Design and Engineering, 7(1):44–55, 2020

    2020

  26. [34]

    Philip Kolvenbach, Oliver Lass, and Stefan Ulbrich. An approach for robust pde-constrained optimization with application to shape optimization of electrical engines and of dynamic elastic structures under uncertainty.Optimization and Engineering, 19:697–731, 2018

    2018

  27. [35]

    Alvaro Maggiar, Andreas Wachter, Irina S Dolinskaya, and Jeremy Staum. A derivative-free trust-region algorithm for the optimization of functions smoothed via gaussian convolution using adaptive multiple importance sampling.SIAM Journal on Optimization, 28(2):1478– 1507, 2018

    2018

  28. [36]

    Optimization and analysis of shock wave/boundary layer interaction for drag reduction by shock control bump.Aerospace Science and Technology, 42:196–208, 2015

    K Mazaheri, KC Kiani, A Nejati, M Zeinalpour, and R Taheri. Optimization and analysis of shock wave/boundary layer interaction for drag reduction by shock control bump.Aerospace Science and Technology, 42:196–208, 2015

    2015

  29. [37]

    Avoiding geometry improvement in derivative-free model-based methods via randomization.arXiv preprint arXiv:2305.17336, 2023

    Matt Menickelly. Avoiding geometry improvement in derivative-free model-based methods via randomization.arXiv preprint arXiv:2305.17336, 2023

    2023 arXiv

  30. [38]

    Robust optimization of large-scale systems.Operations research, 43(2):264–281, 1995

    John M Mulvey, Robert J Vanderbei, and Stavros A Zenios. Robust optimization of large-scale systems.Operations research, 43(2):264–281, 1995

    1995

  31. [39]

    Robust aerodynamic morphing shape optimization for high-lift missions.Aerospace Science and Technology, 103:105897, 2020

    M Nemati and A Jahangirian. Robust aerodynamic morphing shape optimization for high-lift missions.Aerospace Science and Technology, 103:105897, 2020

    2020

  32. [40]

    Random gradient-free minimization of convex functions

    Yurii Nesterov and Vladimir Spokoiny. Random gradient-free minimization of convex functions. Foundations of Computational Mathematics, 17(2):527–566, 2017

    2017

  33. [41]

    Multifidelity approaches for optimization under uncertainty

    Leo WT Ng and Karen E Willcox. Multifidelity approaches for optimization under uncertainty. International Journal for numerical methods in Engineering, 100(10):746–772, 2014

    2014

  34. [42]

    Aerodynamic shape optimization for minimum robust drag and lift reliability constraint.Aerospace Science and Technology, 55:24– 33, 2016

    Dimitrios I Papadimitriou and Costas Papadimitriou. Aerodynamic shape optimization for minimum robust drag and lift reliability constraint.Aerospace Science and Technology, 55:24– 33, 2016

    2016

  35. [43]

    Efficient bilevel surrogate approach for optimization under uncertainty of shock control bumps.AIAA Journal, 58(12):5228–5242, 2020

    Christian Sabater, Philipp Bekemeyer, and Stefan G¨ ortz. Efficient bilevel surrogate approach for optimization under uncertainty of shock control bumps.AIAA Journal, 58(12):5228–5242, 2020

    2020

  36. [44]

    Robust design of transonic natu- ral laminar flow wings under environmental and operational uncertainties.AIAA Journal, 60(2):767–782, 2022

    Christian Sabater, Philipp Bekemeyer, and Stefan G¨ ortz. Robust design of transonic natu- ral laminar flow wings under environmental and operational uncertainties.AIAA Journal, 60(2):767–782, 2022

    2022

  37. [45]

    Surrogate models for uncertainty quantification: An overview

    Bruno Sudret, Stefano Marelli, and Joe Wiart. Surrogate models for uncertainty quantification: An overview. In2017 11th European conference on antennas and propagation (EUCAP), pages 793–797. IEEE, 2017. 23

    2017

  38. [46]

    Application of deep learning based multi-fidelity surrogate model to robust aerodynamic design optimization.Aerospace Science and Technology, 92:722–737, 2019

    Jun Tao and Gang Sun. Application of deep learning based multi-fidelity surrogate model to robust aerodynamic design optimization.Aerospace Science and Technology, 92:722–737, 2019

    2019

  39. [47]

    Impact of turbulence models and shape parameterization on robust aerodynamic shape optimization.Journal of Aircraft, 56(3):1099–1115, 2019

    Aslihan Vuruskan and Serhat Hosder. Impact of turbulence models and shape parameterization on robust aerodynamic shape optimization.Journal of Aircraft, 56(3):1099–1115, 2019

    2019

  40. [48]

    Comparison of full-field and integrated cfd convergence based on richardson extrapolation

    Justin R Weinmeister and Devina P Sanjaya. Comparison of full-field and integrated cfd convergence based on richardson extrapolation. InAIAA A VIATION FORUM AND ASCEND 2024, page 4470, 2024

    2024

  41. [49]

    General robust-optimization formulation for nonlinear programming.Journal of optimization theory and applications, 132(1):111–124, 2007

    Yin Zhang. General robust-optimization formulation for nonlinear programming.Journal of optimization theory and applications, 132(1):111–124, 2007

    2007

  42. [50]

    Yu Zhang, Zhong-hua Han, and Wen-ping Song. An efficient robust aerodynamic design opti- mization method based on a multi-level hierarchical kriging model and multi-fidelity expected improvement.Aerospace Science and Technology, 152:109401, 2024

    2024

  43. [51]

    An inverse design method for airfoils based on pressure gradient distribution.Energies, 13(13):3400, 2020

    Yufei Zhang, Chongyang Yan, and Haixin Chen. An inverse design method for airfoils based on pressure gradient distribution.Energies, 13(13):3400, 2020

    2020

  44. [52]

    Turbulent transonic airfoil flow simulation using a pressure-based algorithm.AIAA Journal, 33(1):42–47, 1995

    Gang Zhou, Lars Davidson, and Erik Olsson. Turbulent transonic airfoil flow simulation using a pressure-based algorithm.AIAA Journal, 33(1):42–47, 1995. 24 Appendix A. Computational Cost of A Multipoint Run The cost of a multipoint run with SLSQP employing adjoint evaluations ...

    1995

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.