Numerical study of non-uniqueness for 2D compressible isentropic Euler equations

Alberto Bressan, Yi Jiang, Hailiang LiuJournal of Computational Physics 2021 · DOI

TL;DR

This paper numerically investigates the long-standing problem of non-uniqueness of solutions for the 2D compressible isentropic Euler equations. By carefully designing initial conditions with singular vorticity, the authors provide compelling numerical evidence that multiple distinct solutions can arise from the same initial data, suggesting a fundamental ill-posedness of the governing equations.

What Problem Does This Paper Address?

The well-posedness of hyperbolic conservation laws, particularly the Euler equations governing fluid dynamics, in multiple spatial dimensions remains a significant open question. While uniqueness of solutions is well-established for one-dimensional problems, the situation in higher dimensions is far more complex. Recent theoretical work has shown the existence of infinitely many weak solutions for the Euler equations, but these solutions are often turbulent and not explicitly constructible. This paper tackles the question of whether physically relevant initial data, specifically those with singular vorticity, can lead to multiple distinct, physically meaningful solutions in the 2D compressible isentropic Euler equations. This is a crucial step towards understanding the fundamental limitations of these equations and the challenges in predicting fluid behavior.

Methods

The authors employ a numerical approach using the positivity-preserving discontinuous Galerkin (DG) method for solving the 2D compressible isentropic Euler equations. The core of their methodology involves:

  • Initial Data Construction: They design initial conditions inspired by previous work on incompressible fluids, featuring vorticity concentrated on two wedges with a singularity at the origin. This vorticity is then approximated by three families of functions (Case 0, 1, and 2) that converge to the singular profile as a perturbation parameter $\epsilon$ goes to zero.
  • Parameter Variation: They systematically vary key parameters in the initial data:
  • $\beta$: the exponent in the initial density profile ($ \rho_0(r, \theta) = r^\beta $).
  • $\alpha$: the exponent in the initial vorticity profile ($ \omega(r, \theta) = r^{-\alpha} \phi(\theta) $).
  • $\theta_0$: the angular support of the initial vorticity.
  • $A$: a parameter in the pressure law ($ p = A\rho^\gamma $), which controls compressibility.
  • Numerical Simulation: The Euler equations are solved using the P¹-DG method in space and a third-order Runge-Kutta method in time, with an invariant-region-preserving limiter.
  • Vorticity Analysis: The primary focus is on analyzing the evolution of the vorticity field to identify differences between solutions originating from the different initial data cases.
  • Metric for Comparison: A metric $D^\epsilon(t)$ is defined to quantify the difference between solutions obtained from Case 0 and Case 2 initial data, helping to assess whether solutions diverge over time.

Datasets

Not discussed in the paper. The paper focuses on numerical simulations generated from carefully constructed initial conditions rather than using pre-existing datasets.

Key Insights and Ideas

  • Numerical Evidence of Non-Uniqueness: The most striking insight is the strong numerical evidence that distinct solutions can emerge from initial data with singular vorticity, even for the compressible Euler equations. This suggests that the problem of non-uniqueness is not confined to incompressible flows.
  • Sensitivity to Initial Data Parameters: The emergence of non-unique solutions is highly sensitive to the specific parameters of the initial data, particularly the exponents $\beta$ and $\alpha$ in the density and vorticity profiles, and the angular support $\theta_0$ of the vorticity.
  • "Incurable" Non-Uniqueness: For certain parameter regimes (e.g., $0 \le \beta < \beta^*$), the numerical results indicate a violation of continuous dependence on initial data, suggesting an "incurable" form of ill-posedness.
  • Role of Compressibility: While the paper primarily focuses on the isentropic case, the exploration of the parameter $A$ (related to compressibility) suggests that less compressible systems tend to form spirals more slowly, but the phenomenon of non-uniqueness persists.
  • Vorticity as a Key Indicator: The vorticity field proves to be a highly effective diagnostic tool for visualizing and distinguishing between different solution branches. The formation of one versus two distinct spirals is a clear qualitative indicator of non-uniqueness.

Key Results

  • Parameter Dependence:
    • For $0 \le \beta < \beta^*$, non-unique solutions are observed, characterized by the formation of two distinct spirals in Case 2, while Case 0 tends to form a single spiral.
  • For $\beta \ge \beta^*$, both cases converge to a single spiral solution.
  • The exponent $\alpha$ in the vorticity profile plays a critical role. For small $\alpha$, solutions are similar. As $\alpha$ increases, non-uniqueness emerges, with Case 0 forming a single spiral and Cases 1 and 2 forming two spirals, before eventually converging to single spirals for larger $\alpha$.
  • The angular support $\theta_0$ of the vorticity is also crucial. For $\theta_0 < \theta_0^*$, non-unique solutions are observed, while for $\theta_0 \ge \theta_0^*$, a single spiral solution is consistently obtained.
  • Convergence of Solutions: The metric $D^\epsilon(t)$ shows an increase over time for parameters leading to non-uniqueness, indicating divergence of solutions, while it remains relatively constant for parameters leading to unique solutions.
  • Mesh Refinement and Asymptotic Behavior: Mesh refinement studies and analysis of the perturbation parameter $\epsilon$ approaching zero confirm the robustness of the observed non-uniqueness.

Limitations and Future Work

  • Numerical Nature: The results are based on numerical simulations. While extensive, they are not rigorous mathematical proofs of the existence of exact solutions. The authors acknowledge that truncation error could play a role, though their findings on refined grids are consistent.
  • Rigorous Validation: A significant future direction is the rigorous validation of these numerical findings, ideally through the construction of exact solutions exhibiting the same non-unique behavior. This would involve developing a posteriori error estimates and potentially computer-assisted proofs.
  • Understanding the Mechanism: While the paper provides strong evidence, a deeper theoretical understanding of the precise mechanism by which these singular initial data lead to multiple solutions is still needed.
  • Generalization: Extending these investigations to three-dimensional flows or more complex initial data remains an open area.

Summary updated

BibTeX
@article{bressan2021numerical,
  author = {Alberto Bressan and Yi Jiang and Hailiang Liu},
  title = {{Numerical study of non-uniqueness for 2D compressible isentropic Euler equations}},
  journal = {Journal of Computational Physics},
  year = {2021},
  doi = {10.1016/j.jcp.2021.110588},
}
Figure