Pith. sign in

REVIEW 3 major objections 6 minor 56 references

Symbolic regression discovers unified skin friction formula from inviscid flow data

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · glm-5.2

2026-07-09 16:10 UTC pith:AIK6F6RW

load-bearing objection Symbolic regression discovers a clean, interpretable skin friction formula chain from Euler surface features — but it's trained on and validated against single-solver RANS/SA data, so the 'discovered physics' may partly encode model artifacts. the 3 major comments →

arxiv 2607.07246 v1 pith:AIK6F6RW submitted 2026-07-08 physics.flu-dyn

Skin friction prediction for attached flows based on two-dimensional inviscid solutions

Mingkun Xia , Shule Zhao , Weiwei Zhang This is my paper
classification physics.flu-dyn
keywords frictionskinanalyticalchaincorrectionflowsphysicalterm
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that a progressive symbolic regression framework, guided by physical reasoning, can discover a single interpretable analytical expression chain for the skin friction coefficient using only inviscid Euler surface solutions (pressure coefficient, local Mach number, streamwise Reynolds number) as inputs. The discovered formula unifies subsonic, supersonic, and hypersonic attached-flow regimes through a hierarchical correction structure: a base logarithmic decay law, a pressure-gradient correction, and a Mach-number correction that transitions from a Prandtl-Glauert compressibility factor to a thermodynamic heating term. The authors validate this formula against RANS and LES data across a range of airfoils, wings, and flow conditions, reporting integrated drag errors of approximately 1-4%.

Core claim

The central object is a four-stage analytical expression chain for the skin friction coefficient c_f, discovered by progressively constraining symbolic regression search spaces. Stage 1 recovers the classical flat-plate turbulent scaling c_f ~ 1/(log(Re_x))^n with n approx 3.18. Stage 2 introduces a multiplicative pressure correction (1 - C_p). Stage 3 couples a Prandtl-Glauert factor sqrt(1 - Ma_x^2) into the pressure term for subsonic compressibility. Stage 4 introduces a sigmoid-activated thermodynamic correction (gamma*Ma_x^2) that activates near Mach 2.5 to model aerodynamic heating effects in supersonic and hypersonic flows. The key structural claim is that compressibility and thermody

What carries the argument

Progressive symbolic regression using PySR with a simple-to-complex search strategy. At each stage, the previously discovered expression structure is frozen and the search space is restricted to correction terms for newly introduced physical effects. Input features (Re, L, C_p, Ma_x) are selected via ensemble learning feature importance (LightGBM, Random Forest) combined with boundary-layer-theory-guided physical reasoning. The operator set includes power laws and logarithmic functions. The loss function is L2 on c_f. The resulting expression chain is validated by integrated skin friction drag relative error against RANS/LES reference solutions.

Load-bearing premise

The method assumes that inviscid Euler surface solutions (pressure coefficient and local Mach number) together with the freestream Reynolds number contain sufficient information to reconstruct viscous skin friction distributions for attached flows, without requiring boundary layer velocity profile data or turbulence intensity. If the internal boundary layer state carries independent information not captured by these surface quantities, the formula chain would not generalize.

What would settle it

A flow case with attached boundary layers but strong favorable or adverse pressure gradients where the surface pressure distribution and local Mach number are similar to a training case, yet the skin friction distribution differs significantly due to boundary layer history effects not encoded in the inviscid surface features. If such a case produces large prediction errors, the premise that inviscid surface features suffice would be falsified.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • The formula chain provides a fast, white-box alternative to full Navier-Stokes simulations for skin friction estimation in attached-flow aerodynamic design optimization, potentially reducing computational cost by orders of magnitude.
  • The progressive discovery framework demonstrates a methodology for extracting physically interpretable scaling laws from simulation data, which could be applied to other aerodynamic quantities such as heat transfer or pressure drag.
  • The finding that inviscid surface features suffice for skin friction prediction in attached flows suggests that coupled Euler-boundary-layer methods could be replaced by direct algebraic surrogates in preliminary design loops.
  • The unified structure across Mach regimes, with natural degeneration to classical forms, provides a template for discovering regime-continuous physical laws in other multi-physics domains.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. This paper proposes a progressive symbolic regression framework to discover an analytical expression chain for skin friction coefficient prediction using inviscid Euler surface solutions (Re_x, C_p, Ma_x) as inputs. The method proceeds in four stages: (1) a base logarithmic law consistent with classical flat-plate theory, (2) a pressure correction term (1-C_p), (3) a subsonic compressibility correction recovering the Prandtl-Glauert factor, and (4) a supersonic/hypersonic thermodynamic correction involving a sigmoid activation. The formula chain is validated against RANS (SA model) solutions across subsonic to hypersonic attached flows, 2D variable-geometry airfoils, and a 3D wing, with reported integrated drag errors of 1-4%. The approach is noteworthy for producing a white-box, interpretable formula rather than a black-box surrogate model.

Significance. The paper makes a genuine contribution to the intersection of data-driven modeling and aerodynamic force prediction. The progressive discovery framework, which freezes expression structure stage-by-stage and searches only for correction terms, is a sensible strategy for constraining the symbolic regression search space and ensuring physical consistency. The recovery of the Prandtl-Glauert factor in the subsonic regime (Eq. 3) and the consistency of the base exponent n≈3.18 with classical Prandtl-Schlichting and Wieghardt formulas (Appendix A) are concrete strengths. The formula chain is concise, falsifiable, and directly usable in engineering design loops. The generalization to unseen geometries (Section 3.2.4) and a 3D wing (Figure 7) is a meaningful test of extrapolation capability.

major comments (3)
  1. Section 3.1, Eq. (4): The supersonic/hypersonic correction term contains a sigmoid factor 1/(1+exp(-0.5(Ma_x-2.5))) with a threshold of 2.5 and a slope parameter of 0.5. These are numerical constants with no derivation from first principles. The paper states this 'indicates that aerodynamic heating gradually takes effect when the Mach number exceeds approximately 2.5,' but this interpretation is post-hoc. Since the formula is both discovered from and validated against SA-model RANS data from a single solver (Section 2.1), this term may encode SA-model-specific response to compressibility rather than universal physics. The authors should either (a) validate Eq. (4) against experimental skin friction data or RANS solutions from a different turbulence model (e.g., k-omega SST) or solver, or (b) substantially soften the claim that the formula chain 'reveals scaling laws' (Abstract, Section 3
  2. Sections 2.1 and 3.2: All training and validation data come from the PhengLEI solver with the SA turbulence model (except one case using the Yang et al. closure). The only independent validation is a single LES comparison at Ma=0.15, Re=4x10^5, alpha=0 (Figure 5a), which shows non-trivial deviations near the leading edge. This is insufficient to establish that the discovered formula chain encodes physical laws rather than model-specific behavior. At minimum, the authors should add validation against RANS solutions from a second turbulence model or solver for at least one subsonic and one supersonic case, or against experimental data if available. Without this, the generalization claims across Mach numbers and geometries are conditioned on SA-model fidelity, which should be stated explicitly in the Abstract and Conclusions.
  3. Section 3.2.3: The exponent n is fine-tuned from 3.18 to approximately 3.28 for low-Reynolds-number conditions (Re~10^5), which are outside the training range (Re~10^6-10^7). This post-hoc recalibration indicates that the base formula does not generalize to Reynolds numbers outside the training space without parameter adjustment. The authors should clarify whether this recalibration was performed using samples from the Pre_lowRe dataset (which is described as a validation set in Table 1), and if so, whether this constitutes a form of data leakage from the test set. The distinction between 'validation' and 'fine-tuning' data should be made explicit.
minor comments (6)
  1. Table 1: The 'SR_Yang' turbulence model is referenced but not clearly defined in the main text; the footnote references Yang et al. [43] but a brief description of how this model differs from standard SA would help readers assess the first training set.
  2. Section 2.2, Figure 2: The feature importance rankings from LightGBM and Random Forest are presented but the specific dataset used for this analysis is not stated. Clarifying which data (Train_1, Train_2, Train_3, or a combination) were used would improve reproducibility of the feature selection.
  3. Abstract: The phrase 'thermodynamic effects term in supersonic and hypersonic regimes' is vague; a brief indication of what thermodynamic effects are captured (e.g., density decrease due to aerodynamic heating) would help readers.
  4. Figure 5a: The LES comparison is valuable but appears at only one condition. If additional LES or DNS data are available for any of the other validation cases, including them would strengthen the independence of the validation.
  5. Section 3.1, Eq. (5): The asymptotic analysis relating C_p to Ma_x is compact but the step from the exact isentropic relation to the proportionality C_p ~ 1/Ma_x^(2*gamma/(gamma-1)) is abrupt; an intermediate step or reference would aid readability.
  6. The paper uses 'log' without specifying the base. Given that the exponent n is calibrated against classical formulas using log10 (Appendix A, Eqs. 14-15), the base should be stated explicitly in Eqs. (1)-(4).

Circularity Check

0 steps flagged

No significant circularity found; the formula chain maps genuinely distinct quantities (inviscid Euler features to viscous c_f) and does not reduce to its inputs by construction

full rationale

The paper's derivation chain is not circular. The symbolic regression discovers a mapping from inviscid surface flow features (C_p, Re_x, Ma_x from Euler solutions) to viscous skin friction c_f (from RANS/SA solutions). These are genuinely different quantities computed from different equation systems, so the output is not equivalent to the input by construction. The formula's structural consistency with classical boundary layer theory is independently corroborated in Appendix A, where an explicit derivation from velocity distribution laws (Eqs. 7-13) yields a form structurally identical to the symbolic regression result, and the exponent n≈3.18 is cross-validated against Prandtl-Schlichting and Wieghardt formulas (n≈3.199). The single-solver/single-turbulence-model concern (all training and most validation data from PhengLEI/SA) is a generalization and external validity risk, not circularity — the validation cases use different geometries, Mach numbers, and Reynolds numbers outside the training space, and one LES comparison is provided. The self-citation to [45] (Xia and Zhang) for the progressive discovery framework is methodological and not load-bearing for the specific physical formulas discovered here. The fitted sigmoid parameters in Eq. 4 are empirical with post-hoc physical interpretation, but this is a justification weakness, not circularity. Score 1 reflects the minor methodological self-citation without load-bearing circularity.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physical entities or particles. The free parameters (n, sigmoid constants) are fitted to data. The axioms are domain assumptions standard in engineering aerodynamics, though the progressive freezing strategy is somewhat ad hoc to this methodology.

free parameters (3)
  • n (base exponent) = 3.18
    Fitted to training data; adjusted to 3.28 for low-Re conditions, indicating it is not a universal constant.
  • sigmoid parameters (0.5, 2.5) = 0.5, 2.5
    The factor 1/(1+exp(-0.5(Ma_x-2.5))) in Eq. 4 contains fitted constants 0.5 and 2.5 that control the transition onset of thermodynamic effects.
  • gamma (specific heat ratio) = 1.4
    Standard physical constant for air, treated as input.
axioms (4)
  • domain assumption Inviscid Euler surface solutions contain sufficient information to reconstruct viscous skin friction for attached flows.
    Invoked in Section 2.2 and the overall methodology; the formula chain uses only C_p, Ma_x, and Re as inputs.
  • domain assumption RANS solutions with the SA turbulence model provide ground-truth skin friction data for training.
    Used throughout Section 2.1 for data construction; the discovered formula inherits SA model assumptions.
  • ad hoc to paper The progressive stage-by-stage freezing of expression structure preserves physical consistency.
    The progressive framework (Section 2) assumes that fixing structure at each stage and searching only correction terms yields physically consistent results; the authors note this strategy influences final results.
  • domain assumption Attached flow assumption: no shock-induced separation or strong nonlinear flow features.
    Stated as a limitation in Section 4; the formula is not validated for transonic flows with shocks.

pith-pipeline@v1.1.0-glm · 22575 in / 2616 out tokens · 342785 ms · 2026-07-09T16:10:04.747087+00:00 · methodology

0 comments
read the original abstract

Boundary layer theory and its analytical methods for skin friction coefficients provide an important basis for aerodynamic analysis. However, classical analytical formulas are mostly limited to flat-plate flows. High-fidelity numerical simulations are not only computationally expensive but also yield predictions that are highly sensitive to physical models, numerical schemes, and grid resolution. To overcome these limitations, symbolic AI opens a new pathway to discover novel laws of complex physical systems from data. Using limited data from surface solutions of the Euler equations and the skin friction coefficient from viscous flows over airfoils, we employ symbolic regression to progressively discover a generalizable, interpretable analytical formula chain for fast skin friction prediction in subsonic and supersonic attached flows. From the perspective of physical mechanisms, the discovered analytical expression chain reveals scaling laws for skin friction at different Mach numbers: the basic form captures the logarithmic decay of skin friction along the streamwise direction in the turbulent boundary layer; the inclusion of a pressure coefficient correction term quantifies the effect of surface pressure variation; and the Mach number correction term evolves with flow regimes, transitioning from the compressibility correction term in subsonic regimes to the thermodynamic effects term in supersonic and hypersonic regimes. This knowledge chain exhibits a unified structure across different Mach numbers, and omitting the correction terms under certain conditions recovers classical theoretical forms, further demonstrating its physical consistency. Validation against typical geometries shows that this analytical formula chain achieves a low average integrated skin friction drag prediction error, with good generalization capability across different freestream conditions and geometric shapes.

Figures

Figures reproduced from arXiv: 2607.07246 by Mingkun Xia, Shule Zhao, Weiwei Zhang.

Figure 1
Figure 1. Figure 1: Workflow diagram of the progressive symbolic regression method. The method is based on [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Feature importance ranking from ensemble learning algorithms. Normalized evaluation results [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of skin friction distribution predictions for low-speed and subsonic conditions. (a) [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Validation of the supersonic formula for skin friction distribution prediction under super [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Validation under low-Reynolds-number conditions (on the order of [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Generalization validation results for two-dimensional variable geometries. (a) The eight new [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Generalization validation results for a three-dimensional wing. Contour plot of the predicted [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗

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

56 extracted references · 56 canonical work pages · 2 internal anchors

  1. [1]

    McGraw hill, 2011

    John Anderson.EBOOK: Fundamentals of Aerodynamics (SI units). McGraw hill, 2011

    work page 2011

  2. [2]

    Airfoil drag at low-to-medium reynolds numbers: A novel estimation method.AIAA journal, 58(7):2791–2805, 2020

    Falk Götten, Marc Havermann, Carsten Braun, Matthew Marino, and Cees Bil. Airfoil drag at low-to-medium reynolds numbers: A novel estimation method.AIAA journal, 58(7):2791–2805, 2020

    work page 2020

  3. [3]

    Free-flight measurements of turbulent-boundary-layer skin friction in the presence of severe aerodynamic heating at mach numbers from 2.8 to 7.0

    Simon C Sommer and Barbara J Short. Free-flight measurements of turbulent-boundary-layer skin friction in the presence of severe aerodynamic heating at mach numbers from 2.8 to 7.0. Technical report, 1955

    work page 1955

  4. [4]

    Springer, 1961

    Hermann Schlichting and Joseph Kestin.Boundary layer theory, volume 121. Springer, 1961

    work page 1961

  5. [5]

    Compressible turbulent boundary layer with pressure gradient and heat transfer.AIAA Journal, 4(1):19–25, 1966

    Philip K Sasman and Robert J Cresci. Compressible turbulent boundary layer with pressure gradient and heat transfer.AIAA Journal, 4(1):19–25, 1966

    work page 1966

  6. [6]

    Engineering calculation method for boundary layer parameters of compressible flows.Chinese Journal of Theoretical and Applied Mechanics, 16(2):120–128, 1980

    Yingui Bian and Guoying Zhao. Engineering calculation method for boundary layer parameters of compressible flows.Chinese Journal of Theoretical and Applied Mechanics, 16(2):120–128, 1980

    work page 1980

  7. [7]

    Precise drag prediction of airfoil flows by a new algebraic model

    Meng-Juan Xiao and Zhen-Su She. Precise drag prediction of airfoil flows by a new algebraic model. Acta Mechanica Sinica, 36(1):35, 2020

    work page 2020

  8. [8]

    On skin friction in wall-bounded turbulence.Acta Mechanica Sinica, 37(4):589–598, 2021

    Zhenhua Xia, Peng Zhang, and Xiang IA Yang. On skin friction in wall-bounded turbulence.Acta Mechanica Sinica, 37(4):589–598, 2021

    work page 2021

  9. [9]

    Reconstructing self-similarity for improving wall skin-friction prediction under adverse pressure gradients.Journal of Fluid Mechanics, 1034:A31, 2026

    Ya-Sen Hu, Zhen-Hua Wan, Peng-Jun-Yi Zhang, De-Jun Sun, and Xi-Yun Lu. Reconstructing self-similarity for improving wall skin-friction prediction under adverse pressure gradients.Journal of Fluid Mechanics, 1034:A31, 2026

    work page 2026

  10. [10]

    Revisitingthetheoryofvandriest: ageneralscalinglawfortheskin-friction coefficient of high-speed turbulent boundary layers.Journal of Fluid Mechanics, 1012:R3, 2025

    ZhiyeZhaoandLinFu. Revisitingthetheoryofvandriest: ageneralscalinglawfortheskin-friction coefficient of high-speed turbulent boundary layers.Journal of Fluid Mechanics, 1012:R3, 2025

    work page 2025

  11. [11]

    Generalisationoftheprandtlrelationfortheskin-frictionlawtocompressible turbulent channel and pipe flows.Journal of Fluid Mechanics, 1032:A72, 2026

    ZhiyeZhaoandLinFu. Generalisationoftheprandtlrelationfortheskin-frictionlawtocompressible turbulent channel and pipe flows.Journal of Fluid Mechanics, 1032:A72, 2026

    work page 2026

  12. [12]

    Numerical evaluation of airfoil friction drag.Journal of aircraft, 37(2):354–356, 2000

    Giovanni Lombardi, Maria Vittoria Salvetti, and Dario Pinelli. Numerical evaluation of airfoil friction drag.Journal of aircraft, 37(2):354–356, 2000

    work page 2000

  13. [13]

    A coupled inviscid–viscous airfoil analysis solver, revisited.AIAA Journal, 60(5):2961–2971, 2022

    Krzysztof J Fidkowski. A coupled inviscid–viscous airfoil analysis solver, revisited.AIAA Journal, 60(5):2961–2971, 2022

    work page 2022

  14. [14]

    Transonic airfoil analysis based on the interaction of euler and integral boundary-layer equation.Acta Aeronautica et Astronautica Sinica, 12(5):221–227, 1991

    Hanlong Cai and Feng Li. Transonic airfoil analysis based on the interaction of euler and integral boundary-layer equation.Acta Aeronautica et Astronautica Sinica, 12(5):221–227, 1991

    work page 1991

  15. [15]

    Numerical simulation of three-dimensional supersonic flows using euler and boundary layer solvers

    F Gaible, R Cariou, and R Houdeville. Numerical simulation of three-dimensional supersonic flows using euler and boundary layer solvers. In31st Aerospace Sciences Meeting, page 531, 1993

    work page 1993

  16. [16]

    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(1):477–508, 2020

    work page 2020

  17. [17]

    Some research progress and prospect of intelligent aerodynamics.Acta Aerodynamica Sinica, 41(7):1–35, 2023

    Zhigong Tang, Linyang Zhu, Xinghao Xiang, Lei HE, Tun ZHAO, Yueqing WANG, Weiqi QIAN, and Xianxu YUAN. Some research progress and prospect of intelligent aerodynamics.Acta Aerodynamica Sinica, 41(7):1–35, 2023

    work page 2023

  18. [18]

    Ascientometricinvestigationofartificialintelligence for fluid mechanics: Emerging topics and active groups.Progress in Aerospace Sciences, 157: 101130, 2025

    WeiweiZhang,MingkunXia,andJiaqingKou. Ascientometricinvestigationofartificialintelligence for fluid mechanics: Emerging topics and active groups.Progress in Aerospace Sciences, 157: 101130, 2025

    work page 2025

  19. [19]

    Envisioning the blueprint: Aeronautics in large models era.Chinese Journal of Aeronautics, 38(8), 2025

    Weiwei Zhang and Shule Zhao. Envisioning the blueprint: Aeronautics in large models era.Chinese Journal of Aeronautics, 38(8), 2025

    work page 2025

  20. [20]

    Fast flow field prediction over airfoils using deep learning approach.Physics of Fluids, 31(5), 2019

    Vinothkumar Sekar, Qinghua Jiang, Chang Shu, and Boo Cheong Khoo. Fast flow field prediction over airfoils using deep learning approach.Physics of Fluids, 31(5), 2019. 18

    work page 2019

  21. [21]

    Goal-orientedfeatureextraction: A novel approach to enhance data-driven surrogate models.AIAA journal, 64(1):303–317, 2026

    XuWang,RuiqiHuang,JiaqingKou,HuiTang,andWeiweiZhang. Goal-orientedfeatureextraction: A novel approach to enhance data-driven surrogate models.AIAA journal, 64(1):303–317, 2026

    work page 2026

  22. [22]

    Fast pressure distribution prediction of airfoils using deep learning.Aerospace Science and Technology, 105:105949, 2020

    Xinyu Hui, Junqiang Bai, Hui Wang, and Yang Zhang. Fast pressure distribution prediction of airfoils using deep learning.Aerospace Science and Technology, 105:105949, 2020

    work page 2020

  23. [23]

    A deep learning framework for aerodynamic pressure prediction on general three-dimensional configurations.Physics of Fluids, 35(10), 2023

    Yang Shen, Wei Huang, Zhen-guo Wang, Da-fu Xu, and Chao-Yang Liu. A deep learning framework for aerodynamic pressure prediction on general three-dimensional configurations.Physics of Fluids, 35(10), 2023

    work page 2023

  24. [24]

    Physics-informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021

    George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. Physics-informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021

    work page 2021

  25. [25]

    Unsteady reduced order model with neural networks and flight-physics-based regularization for aerodynamic applications.Computers & Fluids, 264:105949, 2023

    Mateus Dias Ribeiro, Mario Stradtner, and Philipp Bekemeyer. Unsteady reduced order model with neural networks and flight-physics-based regularization for aerodynamic applications.Computers & Fluids, 264:105949, 2023

    work page 2023

  26. [26]

    Machine learning of skin friction distribution based on surface inviscid flow feature.Chinese Journal of Theoretical and Applied Mechanics, 56(8):2243–2258, 2024

    Shule Zhao and Weiwei Zhang. Machine learning of skin friction distribution based on surface inviscid flow feature.Chinese Journal of Theoretical and Applied Mechanics, 56(8):2243–2258, 2024

    work page 2024

  27. [27]

    Eulerequationembeddedmachinelearningmethodforwallpressure and skin friction distribution.Engineering Applications of Computational Fluid Mechanics, 19(1): 2556450, 2025

    ShuleZhaoandWeiweiZhang. Eulerequationembeddedmachinelearningmethodforwallpressure and skin friction distribution.Engineering Applications of Computational Fluid Mechanics, 19(1): 2556450, 2025

    work page 2025

  28. [28]

    Ed-resnet: Euler equation embedding double-series residual neural network for aerothermal modeling.AIAA Journal, 63(1):134–147, 2025

    Ze Wang, Xu Wang, Weiwei Zhang, and Shufang Song. Ed-resnet: Euler equation embedding double-series residual neural network for aerothermal modeling.AIAA Journal, 63(1):134–147, 2025

    work page 2025

  29. [29]

    Interpretable Machine Learning for Science with PySR and SymbolicRegression.jl

    Miles Cranmer. Interpretable machine learning for science with pysr and symbolicregression. jl. arXiv preprint arXiv:2305.01582, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  30. [30]

    Ai feynman: A physics-inspired method for symbolic regression.Science advances, 6(16):eaay2631, 2020

    Silviu-Marian Udrescu and Max Tegmark. Ai feynman: A physics-inspired method for symbolic regression.Science advances, 6(16):eaay2631, 2020

    work page 2020

  31. [31]

    Hanchen Wang, Tianfan Fu, Yuanqi Du, Wenhao Gao, Kexin Huang, Ziming Liu, Payal Chandak, Shengchao Liu, Peter Van Katwyk, Andreea Deac, Anima Anandkumar, Karianne Bergen, Carla P. Gomes, Shirley Ho, Pushmeet Kohli, Joan Lasenby, Jure Leskovec, Tie-Yan Liu, Arjun Manrai, DeboraMarks,BharathRamsundar,LeSong,JimengSun,JianTang,PetarVeličković,MaxWelling, Lin...

    work page 2023

  32. [32]

    Artificial intelligence in physical sciences: Symbolic regression trends and perspectives: D

    Dimitrios Angelis, Filippos Sofos, and Theodoros E Karakasidis. Artificial intelligence in physical sciences: Symbolic regression trends and perspectives: D. angelis et al.Archives of Computational Methods in Engineering, 30(6):3845–3865, 2023

    work page 2023

  33. [33]

    Distilling free-form natural laws from experimental data.science, 324(5923):81–85, 2009

    Michael Schmidt and Hod Lipson. Distilling free-form natural laws from experimental data.science, 324(5923):81–85, 2009

    work page 2009

  34. [34]

    Data-driven discovery of physical laws.Cognitive Science, 5(1):31–54, 1981

    Pat Langley. Data-driven discovery of physical laws.Cognitive Science, 5(1):31–54, 1981

    work page 1981

  35. [35]

    Harnessingdatausingsymbolicregressionmethodsfordiscovering novel paradigms in physics.Science China Physics, Mechanics & Astronomy, 67(6):267301, 2024

    JianyangGuoandWan-JianYin. Harnessingdatausingsymbolicregressionmethodsfordiscovering novel paradigms in physics.Science China Physics, Mechanics & Astronomy, 67(6):267301, 2024. ISSN 1869-1927. doi: 10.1007/s11433-023-2346-2

    work page doi:10.1007/s11433-023-2346-2 2024

  36. [36]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems.Proceedings of the national academy of sciences, 113(15):3932–3937, 2016

    Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems.Proceedings of the national academy of sciences, 113(15):3932–3937, 2016

    work page 2016

  37. [37]

    Sparseidentificationofmultiphaseturbulence closures for coupled fluid–particle flows.Journal of Fluid Mechanics, 914:A11, 2021

    SarahBeetham,RodneyOFox,andJesseCapecelatro. Sparseidentificationofmultiphaseturbulence closures for coupled fluid–particle flows.Journal of Fluid Mechanics, 914:A11, 2021. 19

    work page 2021

  38. [38]

    Dimensionalhomogeneity constrained gene expression programming for discovering governing equations.Journal of Fluid Mechanics, 985:A12, 2024

    WenjunMa,JunZhang,KaikaiFeng,HaoyunXing,andDongshengWen. Dimensionalhomogeneity constrained gene expression programming for discovering governing equations.Journal of Fluid Mechanics, 985:A12, 2024

    work page 2024

  39. [39]

    Symbolic identification of tensor equations in multidimensional physical fields.Journal of Fluid Mechanics, 1024:A34, 2025

    Tianyi Chen, Hao Yang, Wenjun Ma, and Jun Zhang. Symbolic identification of tensor equations in multidimensional physical fields.Journal of Fluid Mechanics, 1024:A34, 2025

    work page 2025

  40. [40]

    Symbolic regression-enhanced dynamic wake meandering: fast and physically consistent wind turbine wake modelling.Journal of Fluid Mechanics, 1025:A56, 2025

    DingWang,DachuanFeng,KangchengZhou,YuntianChen,Shi-JunLiao,andShiyiChen. Symbolic regression-enhanced dynamic wake meandering: fast and physically consistent wind turbine wake modelling.Journal of Fluid Mechanics, 1025:A56, 2025

    work page 2025

  41. [41]

    A novel evolutionary algorithm applied to algebraic modifications of the rans stress–strain relationship.Journal of Computational Physics, 325:22–37, 2016

    Jack Weatheritt and Richard Sandberg. A novel evolutionary algorithm applied to algebraic modifications of the rans stress–strain relationship.Journal of Computational Physics, 325:22–37, 2016

    work page 2016

  42. [42]

    Data-driven adverse pressure gradient correction for turbulence model.AIAA Journal, 63(7):2780–2796, 2025

    XiangLin Shan and WeiWei Zhang. Data-driven adverse pressure gradient correction for turbulence model.AIAA Journal, 63(7):2780–2796, 2025

    work page 2025

  43. [43]

    Data-enabled discovery of specific and generalisable turbulence closures.Journal of Fluid Mechanics, 1016:R1, 2025

    Zhongxin Yang, Xianglin Shan, Xiang IA Yang, and Weiwei Zhang. Data-enabled discovery of specific and generalisable turbulence closures.Journal of Fluid Mechanics, 1016:R1, 2025

    work page 2025

  44. [44]

    Scaling function learning: A sparse aerodynamic data reconstruction method for various aircraft shapes.Chinese Journal of Aeronautics, 39(4):103797, 2025

    Haitao Lin, Xu Wang, and Weiwei Zhang. Scaling function learning: A sparse aerodynamic data reconstruction method for various aircraft shapes.Chinese Journal of Aeronautics, 39(4):103797, 2025

    work page 2025

  45. [45]

    Data-driven Progressive Discovery of Physical Laws

    MingkunXiaandWeiweiZhang. Data-drivenprogressivediscoveryofphysicallaws.arXivpreprint arXiv:2603.13727, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  46. [46]

    Data-driven discovery of dimensionless numbers and governing laws from scarce measurements.Nature communications, 13(1):7562, 2022

    Xiaoyu Xie, Arash Samaei, Jiachen Guo, Wing Kam Liu, and Zhengtao Gan. Data-driven discovery of dimensionless numbers and governing laws from scarce measurements.Nature communications, 13(1):7562, 2022

    work page 2022

  47. [47]

    Mingkun Xia, Haitao Lin, and Weiwei Zhang. Hierarchical dimensionless learning: A physics- data hybrid-driven approach for discovering dimensionless parameter combinations.Engineering Applications of Artificial Intelligence, 170:114190, 2026

    work page 2026

  48. [48]

    Transformations of velocity and temperature for compressible turbulent boundary layers from subsonic to hypersonic regimes.Journal of Fluid Mechanics, 1029:A12, 2026

    Xiang Zhang, Yuntian Chen, Zi-Mo Liao, Zhiye Zhao, Chenyue Xie, Nan-Sheng Liu, and Xi-Yun Lu. Transformations of velocity and temperature for compressible turbulent boundary layers from subsonic to hypersonic regimes.Journal of Fluid Mechanics, 1029:A12, 2026

    work page 2026

  49. [49]

    Extracting self-similarity from data

    Nikos Bempedelis, Luca Magri, and Konstantinos Steiros. Extracting self-similarity from data. Journal of Fluid Mechanics, 1020:A11, 2025

    work page 2025

  50. [50]

    Symbolic learning of material constitutive laws.Science China Physics, Mechanics & Astronomy, 69(1), 2025

    Yang Liu and Jingdi Wan. Symbolic learning of material constitutive laws.Science China Physics, Mechanics & Astronomy, 69(1), 2025

    work page 2025

  51. [51]

    Design and development of software automated continuous integration and testing platform for national numerical windtunnel project.Acta Aerodynamica Sinica, 38(6):1158–1164, 2021

    Lei He, Yongyan Guo, Zhichun Zeng, Xiangcheng Lai, and Zhong Zhao. Design and development of software automated continuous integration and testing platform for national numerical windtunnel project.Acta Aerodynamica Sinica, 38(6):1158–1164, 2021

    work page 2021

  52. [52]

    Largesamplepropertiesofsimulationsusinglatinhypercubesampling.Technometrics, 29(2):143–151, 1987

    MichaelStein. Largesamplepropertiesofsimulationsusinglatinhypercubesampling.Technometrics, 29(2):143–151, 1987

    work page 1987

  53. [53]

    Lightgbm: A highly efficient gradient boosting decision tree.Advances in neural information processing systems, 30, 2017

    Guolin Ke, Qi Meng, Thomas Finley, TaifengWang, WeiChen, Weidong Ma, Qiwei Ye, and Tie-Yan Liu. Lightgbm: A highly efficient gradient boosting decision tree.Advances in neural information processing systems, 30, 2017

    work page 2017

  54. [54]

    Random forests.Machine learning, 45(1):5–32, 2001

    Leo Breiman. Random forests.Machine learning, 45(1):5–32, 2001

    work page 2001

  55. [55]

    McGraw hill, 2011

    FM White, C Ng, and S Saimek.Fluid Mechanics. McGraw hill, 2011

    work page 2011

  56. [56]

    An explicit expression of the turbulent frictional drag coefficient for a flat plate

    Changchun Shi. An explicit expression of the turbulent frictional drag coefficient for a flat plate. Acta Aeronautica et Astronautica Sinica, 9:B299–B302, 1988. 20

    work page 1988