pith. machine review for the scientific record. sign in

arxiv: 2604.14567 · v1 · submitted 2026-04-16 · ⚛️ physics.flu-dyn

Recognition: unknown

A Discrete Adjoint Gas-Kinetic Scheme for Aerodynamic Shape Optimization in Turbulent Continuum Flows

Hangkong Wu , Yuze Zhu , Yajun Zhu , Kun Xu
Authors on Pith no claims yet

Pith reviewed 2026-05-10 10:41 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords discrete adjointgas-kinetic schemeaerodynamic shape optimizationturbulent flowsalgorithmic differentiationSpalart-Allmaras modelsensitivity analysiscontinuum flows
0
0 comments X

The pith

A backward-mode discrete adjoint gas-kinetic scheme produces sensitivities that match a forward linearized solver and supports efficient shape optimization in turbulent continuum flows.

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

The paper develops a discrete adjoint version of the gas-kinetic scheme for computing sensitivities in aerodynamic flows. It uses backward algorithmic differentiation to build the adjoint and verifies it by showing that the sensitivities agree closely with those from a forward-mode linearized version of the same scheme. The method is then applied to optimize shapes under the Spalart-Allmaras turbulence model in three test cases involving turbine blades and airfoils. A sympathetic reader would care because accurate and efficient sensitivity information allows gradient-based optimization to improve designs like lift-to-drag ratios without excessive computational cost.

Core claim

The central claim is that a discrete adjoint gas-kinetic scheme, constructed using the backward mode of algorithmic differentiation, provides accurate sensitivities for shape optimization in turbulent continuum flows. This is established by rigorous verification against a duality-preserving linearized gas-kinetic scheme generated via forward-mode algorithmic differentiation, showing matching convergence behaviors and negligible discrepancies. The approach is demonstrated in fully turbulent optimizations using the Spalart-Allmaras model for inverse design of turbine blades, enhancement of lift-to-drag ratio, and reduction of shock strength on a NACA 0012 airfoil, where design objectives are 0

What carries the argument

The discrete adjoint solver for the gas-kinetic scheme obtained through backward-mode algorithmic differentiation, which ensures duality preservation and exact matching of sensitivities with the linearized forward solver.

Load-bearing premise

That the backward-mode algorithmic differentiation applied to the gas-kinetic scheme produces a duality-preserving adjoint whose sensitivities exactly match those obtained from the forward-mode linearized solver.

What would settle it

If a direct comparison of sensitivity values from the discrete adjoint and the linearized solver revealed discrepancies larger than numerical round-off error in the benchmark cases, or if the shape optimizations failed to improve the design objectives despite the predicted sensitivities.

Figures

Figures reproduced from arXiv: 2604.14567 by Hangkong Wu, Kun Xu, Yajun Zhu, Yuze Zhu.

Figure 1
Figure 1. Figure 1: The flow chart of different forms of adjoint methods [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: shows the data flow of the flow GKS solver, which consists of seven main subroutines. The functionality of each part is given by 1. INIT: initialize macroscopic variables Qi and Qb; 2. BC: apply boundary conditions; 3. DF: compute the distribution function f (Eq. 12); 4. SA: compute the eddy viscosity using the SA turbulence model; 5. RES: evaluate the residual R in each control volume; 6. UPDATE: update f… view at source ↗
Figure 3
Figure 3. Figure 3: The dataflow of the first part of the adjoint GKS solver: adjoint GKS equation [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The dataflow of the second part of the adjoint GKS solver: sensitivity evaluation [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The computational mesh for the Durham turbine cascade case [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distributions of p and Cp for the Durham turbine cascade case: a) in the whole computational domain; b) on blade surfaces (Exp in the legend represents the experimental data) 5.1.2. NACA 0012 Airfoil The second validation case for the flow GKS solver is the NACA 0012 air￾foil [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The computational mesh of the NACA 0012 airfoil [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The distributions of p and Cp for the NACA 0012 airfoil: a)in the whole computa￾tional domain; b) on airfoil surface (a) flow fields (b) adjoint fields [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Both the flow and adjoint variable contours related to the turbulence model equation [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Evolutionary histories of sensitivities and RMS residuals for both adjoint and [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Both the flow and adjoint variable contours related to the turbulence model equa [PITH_FULL_IMAGE:figures/full_fig_p021_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Evolutionary histories of sensitivities and RMS residuals for both adjoint and [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The dataflow of the adjoint-based design optimization system [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: presents the distribution of five sets of Hicks-Henne hump func￾tions. It can be seen that the Hicks-Henne hump function has very obvious localization property [PITH_FULL_IMAGE:figures/full_fig_p023_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Comparison of mesh quality between the original and optimized airfoils. [PITH_FULL_IMAGE:figures/full_fig_p024_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Evolutionary history of the objective function for the Durham turbine cascade case [PITH_FULL_IMAGE:figures/full_fig_p025_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Comparison of the Mach number distributions on blade surfaces and blade profiles [PITH_FULL_IMAGE:figures/full_fig_p026_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Evolutionary history of the objective function for the NACA 0012 airfoil [PITH_FULL_IMAGE:figures/full_fig_p027_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Comparison of original and optimized NACA 0012 airfoils [PITH_FULL_IMAGE:figures/full_fig_p027_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Comparison of the static pressure contours between the original and optimized [PITH_FULL_IMAGE:figures/full_fig_p028_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: Comparison of the static pressure on surface between the original and optimized [PITH_FULL_IMAGE:figures/full_fig_p028_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Entropy contours for the NACA 0012 airfoil: a) viscous flow; b) inviscid flow [PITH_FULL_IMAGE:figures/full_fig_p030_22.png] view at source ↗
Figure 23
Figure 23. Figure 23: Evolutionary history of the objective function for reducing shock strength of the [PITH_FULL_IMAGE:figures/full_fig_p030_23.png] view at source ↗
Figure 24
Figure 24. Figure 24: The entropy contours for the optimized NACA 0012 airfoil [PITH_FULL_IMAGE:figures/full_fig_p031_24.png] view at source ↗
Figure 25
Figure 25. Figure 25: Comparison of the Mach number contours between the original and optimized [PITH_FULL_IMAGE:figures/full_fig_p031_25.png] view at source ↗
Figure 26
Figure 26. Figure 26: Comparison of the static pressure on the blade surface between the original and [PITH_FULL_IMAGE:figures/full_fig_p031_26.png] view at source ↗
Figure 27
Figure 27. Figure 27: Comparison of the original and optimized NACA 0012 airfoils for the shock-strength [PITH_FULL_IMAGE:figures/full_fig_p032_27.png] view at source ↗
Figure 28
Figure 28. Figure 28: The dataflow of the linearized GKS solver [PITH_FULL_IMAGE:figures/full_fig_p033_28.png] view at source ↗
read the original abstract

This study presents an efficient and accurate discrete adjoint gas-kinetic scheme (GKS) for sensitivity analysis and aerodynamic shape optimization in continuum flow regimes. Developed using the backward mode of algorithmic differentiation (AD), the adjoint solver is rigorously verified against a duality-preserving linearized GKS solver generated via forward-mode AD. The robustness and practical effectiveness of the solver are evaluated through three benchmark cases: the inverse design of turbine blades, lift-to-drag ratio enhancement, and shock-strength reduction for a NACA 0012 airfoil. To capture realistic flow physics, fully turbulent optimizations are conducted using the one-equation Spalart--Allmaras (SA) model. Numerical results demonstrate excellent agreement between the discrete adjoint and linearized solvers, exhibiting matching sensitivity convergence behaviors, identical asymptotic residual decay rates, and negligible discrepancies in final sensitivity predictions. Furthermore, the optimization studies confirm that targeted design objectives are consistently achieved within a limited number of design cycles, highlighting the solver's computational efficiency, accuracy, and suitability for complex aerodynamic geometries.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 3 minor

Summary. The paper develops a discrete adjoint gas-kinetic scheme (GKS) for sensitivity analysis and aerodynamic shape optimization in turbulent continuum flows. The adjoint is constructed via backward-mode algorithmic differentiation and verified against a forward-mode linearized GKS solver, with the verification showing matching sensitivity convergence, residual decay rates, and final values. The method is demonstrated on three fully turbulent benchmark optimizations under the Spalart-Allmaras model: inverse design of turbine blades, lift-to-drag ratio maximization, and shock-strength reduction on a NACA 0012 airfoil. Results indicate that design objectives are achieved in a limited number of cycles.

Significance. If the reported numerical agreement holds, the work supplies a duality-preserving discrete adjoint for GKS that enables reliable gradient-based optimization in turbulent regimes. The use of algorithmic differentiation guarantees consistency between forward and adjoint operators, which is a strength for complex discretizations. The three benchmark cases provide concrete evidence of practical utility for aerodynamic design problems involving turbulence.

minor comments (3)
  1. The abstract states 'excellent agreement' and 'negligible discrepancies' but does not report quantitative measures such as L2 norms of sensitivity differences or maximum relative errors; adding these would strengthen the verification claim without altering the central result.
  2. Section 4 (or equivalent verification section) should explicitly state the mesh resolutions and iteration counts used for the sensitivity comparisons to allow direct reproduction of the reported residual decay rates.
  3. Figure captions for the optimization histories could include the number of design variables and the final objective improvement percentages for each case.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our work on the discrete adjoint gas-kinetic scheme and for recommending minor revision. The recognition of the method's consistency via algorithmic differentiation and its utility for turbulent aerodynamic optimization is appreciated. No specific major comments were provided in the report.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper derives a discrete adjoint GKS via backward-mode AD and verifies it directly against an independently implemented forward-mode linearized GKS solver, showing matching convergence, residual decay, and sensitivity values. This cross-check is external to the adjoint construction itself and does not reduce any claimed result to a tautology or fitted input. No load-bearing steps invoke self-citations, uniqueness theorems from prior author work, or ansatzes that loop back to the target sensitivities; the SA turbulence model and benchmark optimizations are standard and independently falsifiable. The derivation chain remains self-contained against the provided external verification benchmark.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

No explicit free parameters, invented entities, or ad-hoc axioms are stated in the abstract. The work implicitly rests on standard CFD assumptions about the validity of the gas-kinetic scheme in continuum regimes and the adequacy of the Spalart-Allmaras model.

axioms (1)
  • domain assumption The gas-kinetic scheme accurately represents continuum turbulent flows when coupled with the Spalart-Allmaras model.
    Invoked for all benchmark optimizations without additional justification visible in the abstract.

pith-pipeline@v0.9.0 · 5482 in / 1308 out tokens · 35029 ms · 2026-05-10T10:41:48.477994+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

31 extracted references · 28 canonical work pages

  1. [1]

    Z. Han, J. Qiao, L. Zhang, Q. Chen, H. Yang, Y. Ding, K. Zhang, W. Song, B. Song, Recent progress of efficient low-boom design and optimization methods, Progress in Aerospace Sciences 146 (2024) 101007.doi:doi. org/10.1016/j.paerosci.2024.101007

    work page doi:10.1016/j.paerosci.2024.101007 2024

  2. [2]

    Zhang, Z

    Y. Zhang, Z. H. Han, W. P. Song, Multi-fidelity expected improvement based on multi-level hierarchical Kriging model for efficient aerodynamic design optimization, Engineering Optimization 56 (12) (2024) 2408–2430. doi:doi.org/10.1080/0305215X.2024.2310182. 34

    work page doi:10.1080/0305215x.2024.2310182 2024

  3. [3]

    J. Li, J. Luo, Z. Fu, F. Liu, Adjoint-based aerodynamic design optimiza- tion of a compressor rotor blade considering static elastic deformation, Aerospace Science and Technology 168 (2026) 110842.doi:doi.org/10. 1016/j.ast.2025.110842

    work page arXiv 2026

  4. [4]

    S. Xu, D. Radford, M. Meyer, J.-D. Müller, Stabilisation of discrete steady adjointsolvers, JournalofComputationalPhysics299(2015)175–195.doi: doi.org/10.1016/j.jcp.2015.06.036

    work page doi:10.1016/j.jcp.2015.06.036 2015

  5. [5]

    Jameson, Aerodynamic design via control theory, Journal of Scientific Computing 3 (1988) 233–260.doi:doi.org/10.1007/BF01061285

    A. Jameson, Aerodynamic design via control theory, Journal of Scientific Computing 3 (1988) 233–260.doi:doi.org/10.1007/BF01061285

    work page doi:10.1007/bf01061285 1988

  6. [6]

    Jameson, L

    A. Jameson, L. Martinelli, N. Pierce, Optimum aerodynamic design using the Navier–Stokes equations, Theoretical and Computational Fluid Dy- namics 10 (1998) 213–237.doi:doi.org/10.1007/s001620050060

    work page doi:10.1007/s001620050060 1998

  7. [7]

    Saurabh, D

    S. Vitale, M. Pini, P. Colonna, Multistage turbomachinery design using the discrete adjoint method within the open-source software SU2, Journal of Propulsion and Power 36 (3) (2020) 465–478.doi:doi.org/10.2514/1. B37685

    work page doi:10.2514/1 2020

  8. [8]

    H. Wu, X. Da, D. Wang, X. Huang, Multi-row turbomachinery aerodynamic design optimization by an efficient and accurate dis- crete adjoint solver, Aerospace 10 (2) (2023).doi:doi.org/10.3390/ aerospace10020106

    2023

  9. [9]

    C. Ma, X. Su, X. Yuan, An efficient unsteady adjoint optimization system for multistage turbomachinery, Journal of Turbomachinery 139 (1) (2016) 011003.doi:10.1115/1.4034185

    work page doi:10.1115/1.4034185 2016

  10. [10]

    Rubino, S

    A. Rubino, S. Vitale, P. Colonna, M. Pini, Fully-turbulent adjoint method for the unsteady shape optimization of multi-row turbomachin- ery, Aerospace Science and Technology 106 (2020) 106132.doi:doi.org/ 10.1016/j.ast.2020.106132. 35

    work page doi:10.1016/j.ast.2020.106132 2020

  11. [11]

    H. Wu, D. Wang, X. Huang, Three-dimensional unsteady aerodynamic optimization of a transonic compressor stage using an implicit discrete ad- joint harmonic balance method, Journal of Turbomachinery 147 (7) (2024) 071016.doi:10.1115/1.4067290

    work page doi:10.1115/1.4067290 2024

  12. [12]

    L. He, D. X. Wang, Concurrent blade aerodynamic-aero-elastic design opti- mization using adjoint method, Journal of Turbomachinery 133 (1) (2010) 011021.doi:10.1115/1.4000544

    work page doi:10.1115/1.4000544 2010

  13. [13]

    M. C. Duta, M. S. Campobasso, M. B. Giles, L. B. Lapworth, Adjoint harmonic sensitivities for forced response minimization, Journal of En- gineering for Gas Turbines and Power 128 (1) (2003) 183–189.doi: 10.1115/1.2031227

    work page doi:10.1115/1.2031227 2003

  14. [14]

    H. Wu, D. Wang, X. Huang, Multi-objective aerodynamic and aeroelastic coupled design optimization using a full viscosity discrete adjoint harmonic balance method, Journal of Turbomachinery 145 (10) (2023) 101002.doi: 10.1115/1.4062803

    work page doi:10.1115/1.4062803 2023

  15. [15]

    H. Wu, M. Yang, D. Wang, X. Huang, Efficient forced response minimiza- tion using a full-viscosity discrete adjoint harmonic balance method, AIAA Journal 62 (10) (2024) 3644–3661.doi:10.2514/1.J063749

    work page doi:10.2514/1.j063749 2024

  16. [16]

    Nadarajah, A

    S. Nadarajah, A. Jameson, A comparison of the continuous and discrete ad- joint approach to automatic aerodynamic optimization, in: 38th Aerospace Sciences Meeting and Exhibit, 2000.doi:10.2514/6.2000-667

    work page doi:10.2514/6.2000-667 2000

  17. [17]

    Hascoet, V

    L. Hascoet, V. Pascual, The Tapenade automatic differentiation tool: prin- ciples, model, and specification 39 (3) (5 2013).doi:10.1145/2450153. 2450158

    work page doi:10.1145/2450153 2013

  18. [18]

    M. B. Giles, M. C. Duta, J.-D. Muller, N. A. Pierce, Algorithm develop- ments for discrete adjoint methods, AIAA Journal 41 (2) (2003) 198–205. doi:10.2514/2.1961. 36

    work page doi:10.2514/2.1961 2003

  19. [19]

    D. X. Wang, L. He, Adjoint aerodynamic design optimization for blades in multistage turbomachines—part I: methodology and verification, Journal of Turbomachinery 132 (2) (2010) 021011.doi:10.1115/1.3072498

    work page doi:10.1115/1.3072498 2010

  20. [20]

    Roe, Approximate Riemann solvers, parameter vectors, and difference schemes, Journal of Computational Physics 135 (2) (1997) 250–258.doi: doi.org/10.1006/jcph.1997.5705

    P. Roe, Approximate Riemann solvers, parameter vectors, and difference schemes, Journal of Computational Physics 135 (2) (1997) 250–258.doi: doi.org/10.1006/jcph.1997.5705

    work page doi:10.1006/jcph.1997.5705 1997

  21. [21]

    K. Xu, K. H. Prendergast, Numerical Navier-Stokes solutions from gas kinetic theory, Journal of Computational Physics 114 (1) (1994) 9–17.doi: doi.org/10.1006/jcph.1994.1145

    work page doi:10.1006/jcph.1994.1145 1994

  22. [22]

    K. Xu, A gas-kinetic BGK scheme for the Navier–Stokes equations and its connection with artificial dissipation and Godunov method, Journal of Computational Physics 171 (1) (2001) 289–335.doi:doi.org/10.1006/ jcph.2001.6790

    work page arXiv 2001

  23. [24]

    F. Zhao, X. Ji, W. Shyy, K. Xu, A compact high-order gas-kinetic scheme on unstructured mesh for acoustic and shock wave computations, Journal of Computational Physics 449 (2022) 110812.doi:doi.org/10.1016/j. jcp.2021.110812

    work page doi:10.1016/j 2022

  24. [25]

    X. Ji, F. Zhao, W. Shyy, K. Xu, Compact high-order gas-kinetic scheme for three-dimensional flow simulations, AIAA Journal 59 (8) (2021) 2979–2996. doi:10.2514/1.J060208

    work page doi:10.2514/1.j060208 2021

  25. [26]

    L. Pan, K. Xu, Q. Li, J. Li, An efficient and accurate two-stage fourth-order gas-kinetic scheme for the Euler and Navier–Stokes equations, Journal of Computational Physics 326 (2016) 197–221.doi:https://doi.org/10. 1016/j.jcp.2016.08.054. 37

    2016

  26. [27]

    X. Ji, L. Pan, W. Shyy, K. Xu, A compact fourth-order gas-kinetic scheme for the Euler and Navier–Stokes equations, Journal of Computational Physics 372 (2018) 446–472.doi:doi.org/10.1016/j.jcp.2018.06.034

    work page doi:10.1016/j.jcp.2018.06.034 2018

  27. [28]

    R. Yuan, L. Wu, A design optimization method for rarefied and continuum gas flows, Journal of Computational Physics 517 (2024) 113366.doi:doi. org/10.1016/j.jcp.2024.113366

    work page doi:10.1016/j.jcp.2024.113366 2024

  28. [29]

    R. Yuan, L. Wu, Adjoint shape optimization from the continuum to free- molecular gas flows, Journal of Computational Physics 537 (2025) 114102. doi:doi.org/10.1016/j.jcp.2025.114102

    work page doi:10.1016/j.jcp.2025.114102 2025

  29. [30]

    R. Yuan, C. Zhong, A conservative implicit scheme for steady state solu- tions of diatomic gas flow in all flow regimes, Computer Physics Commu- nications 247 (2020) 106972.doi:doi.org/10.1016/j.cpc.2019.106972

    work page doi:10.1016/j.cpc.2019.106972 2020

  30. [31]

    Spalart, S

    P. Spalart, S. Allmaras, A one-equation turbulence model for aerodynamic flows, in: 30th Aerospace Sciences Meeting and Exhibit, 1992.doi:10. 2514/6.1992-439

    1992

  31. [32]

    H. Wu, S. Xu, X. Huang, D. Wang, The development and verification of a fully turbulent discrete adjoint solver using algorithmic differentiation, Vol. Volume 2C: Turbomachinery — Design Methods and CFD Modeling for Turbomachinery; Ducts, Noise, and Component Interactions of Turbo Expo, 2021, p. V02CT34A031.doi:10.1115/GT2021-59610. 38

    work page doi:10.1115/gt2021-59610 2021