The Curious Case Of A NACA 0012 Airfoil: Are We Learning Something New?

Rasha AlJahdali, Lisandro Dalcin, Gianmarco Mengaldo, Matteo ParsaniProceedings of the ... International Conference on Fluid Flow, Heat and Mass Transfer 2025 · DOI

TL;DR

For decades, the NACA 0012 airfoil at a Reynolds number of 5,000 has been a "must-pass" benchmark for steady-state laminar flow solvers. This paper reveals a startling discrepancy: while traditional acceleration techniques force the simulation into a steady state, modern high-fidelity, time-accurate methods suggest the flow is actually unsteady.

What Problem Does This Paper Address?

In Computational Fluid Dynamics (CFD), researchers rely on standard "test cases" to verify that their code is mathematically correct. One of the most famous is the 2D NACA 0012 airfoil under specific conditions ($Re = 5,000, Ma = 0.5, AoA = 0^\circ$). Since 1985, the consensus in the literature has been that this flow is perfectly steady and laminar.

However, most researchers reach this "steady" solution using numerical "tricks"—acceleration algorithms like multigrid methods or large pseudo-time steps designed to kill oscillations and reach a result quickly. The authors of this paper ask a provocative question: Are these numerical smoothers actually suppressing real physical unsteadiness? If we stop trying to "force" a steady state and instead use high-fidelity, time-accurate simulations, what does the physics actually tell us?

Methods

The researchers employed high-order accurate, entropy-stable spatial discretizations. Specifically, they used the SSDC (Streamline-Upwind/Petrov-Galerkin) framework, though they noted consistent results using other high-fidelity solvers like PyFR and Flexi.

Key aspects of their numerical approach include:

  • Spatial Discretization: Polynomial degrees ($deg$) ranging from 1 to 4 (corresponding to 2nd through 5th-order accuracy).
  • Time Advancement: They compared the 3BS scheme (an explicit Runge-Kutta pair) against Backward Euler (BE).
  • Grids: Four different structured meshes (quadrangles and triangles) generated via Gmsh, refined near the trailing edge and wake to capture fine-scale physics.
  • Boundary Conditions: Solid wall no-slip conditions on the airfoil and a "sponge zone" at the outlet to prevent artificial reflections.

Datasets

The study uses the classic NACA 0012 airfoil benchmark. While not a "dataset" in the machine learning sense, it is a foundational benchmark in fluid mechanics. The specific parameters used are:

  • Reynolds Number ($Re$): 5,000 (Laminar regime)
  • Mach Number ($Ma$): 0.5 (Compressible)
  • Prandtl Number ($Pr$): 0.72
  • Angle of Attack ($AoA$): $0^\circ$

Key Insights and Ideas

  • The "Steady" Trap: When using the Backward Euler scheme with a large time step (a common practice to find steady solutions), the residual drops to machine precision ($10^{-12}$), appearing perfectly converged. This matches the "steady" results published for the last 40 years.
  • High-Order Sensitivity: Low-order methods ($deg=1$) tend to stagnate and fail to show unsteadiness. However, as soon as the polynomial degree is increased ($deg \ge 2$), the solver becomes sensitive enough to "see" instabilities in the wake.
  • The Reversion to Unsteadiness: The most "aha" moment occurs when the authors take the "perfectly converged" steady solution and restart the simulation using a small, time-accurate step or an explicit scheme. The residual immediately climbs back up to $O(1)$, and the wake becomes visibly unstable.
  • Numerical vs. Physical Stability: The paper suggests that the "steady state" found in previous literature might be a numerical artifact—a result of using solvers that act as low-pass filters, effectively "ironing out" the physical oscillations that want to occur.

Key Results

  • Residual Behavior: For $deg \ge 2$, the residual initially decreases but then rapidly increases and plateaus at a high value ($O(1)$), indicating a non-steady, oscillating wake.
  • Lift and Drag: In the unsteady state, the lift coefficient ($C_l$) is not zero but oscillates symmetrically around the horizontal axis. While the variations are small, they are persistent and time-dependent.
  • Consistency Across Solvers: Although the paper focuses on the SSDC solver, the authors state that the same "unexpected" unsteadiness was observed across three different high-fidelity frameworks (SSDC, PyFR, and Flexi).

Limitations and Future Work

The authors acknowledge that this is an "ongoing study." While they have demonstrated that the flow appears unsteady with high-fidelity methods, they stop short of a definitive declaration that the 40-year-old consensus is flat-out wrong.

Future work will likely involve:

  • Grid Convergence Studies: Determining if further mesh refinement changes the frequency or magnitude of the observed unsteadiness.
  • Stability Analysis: Performing a formal linear stability analysis to see if the flow is theoretically on the verge of a Hopf bifurcation at these parameters.
  • Broader Verification: Investigating if other "standard" steady test cases exhibit similar hidden unsteadiness when subjected to time-accurate, high-order analysis.

Summary updated

Abstract

Numerical methods and "history-contextualized" high-performance computing have been the cornerstone of computational fluid dynamics (CFD) for the past 70 years, allowing solving of more and more complex models governing flow physics.In CFD, we are at a juncture where it has become possible to use so-called high-fidelity methods to simulate accurate complex flow models, allegedly with high accuracy.The primary method for building and quantifying confidence in modeling and simulation is to verify and validate computational tools.The focus of this paper is on a widely-used test case used for the verification of compressible flow solvers.We report the results for the compressible flow past a 2D NACA 0012 airfoil at Reynolds, Mach, and Prandtl numbers of = 5,000, = 0.5, = 0.72, at zero angle of attack.This test case is reported to be (laminar) steady in the literature.Is that the case?Is this test case teaching us something new about the large class of numerical discretizations used in this work?
BibTeX
@inproceedings{aljahdali2025curious,
  author = {Rasha AlJahdali and Lisandro Dalcin and Gianmarco Mengaldo and Matteo Parsani},
  title = {{The Curious Case Of A NACA 0012 Airfoil: Are We Learning Something New?}},
  booktitle = {Proceedings of the ... International Conference on Fluid Flow, Heat and Mass Transfer},
  year = {2025},
  doi = {10.11159/ffhmt25.146},
}
Figure