Global Ill-Posedness with Bounded Data

• The paper demonstrates norm inflation, showing that small bounded initial data in critical spaces can evolve into arbitrarily large norms in very short times.
• It establishes nonuniqueness of weak solutions for compressible Euler systems using convex integration and oscillatory corrections under energy constraints.
• The study underscores how critical space thresholds and nonlinear resonances bridge the gap between well-posed and ill-posed regimes, challenging classical PDE theory.

Global ill-posedness with bounded initial datum refers to the breakdown of well-posedness (uniqueness and/or continuous dependence on initial data) for evolution partial differential equations (PDEs) even when the initial data are smooth and uniformly bounded in strong function space topologies (such as $L^\infty$, $W^{1,\infty}$, or critical Besov spaces). This phenomenon often manifests as "norm inflation"—arbitrarily small bounded data evolving instantaneously to become arbitrarily large in a critical norm—or as nonuniqueness of admissible weak solutions. Recent research establishes this pathology in a broad array of nonlinear dispersive and hyperbolic PDEs, highlighting the limitations of PDE theory at or below critical regularity and the subtle interplay of nonlinearity, dispersion, and geometry.

1. Ill-posedness via Norm Inflation in Evolution Equations

Norm inflation is a robust mechanism certifying ill-posedness: for any $\epsilon>0$, one constructs smooth, bounded initial data $u_0$ with small norm in a critical space such that the solution exhibits uncontrollable norm growth within arbitrarily short time, violating Hadamard continuity. In the context of the fifth-order Camassa–Holm (FOCH) equation, precise analytic constructions in Besov spaces $B^1_{\infty,1}$ and $B^{3/2}_{2,q}$ for $b=5/3$ exhibit this phenomenon.

For $u_0\in C^{\infty}(\mathbb{R})$ with $\|u_0\|_{B^1_{\infty,1}}\le C N^{-1/10}$, the unique smooth solution $u(t,x)$ develops $\|u(t_0)\|_{B^1_{\infty,1}} \ge \ln N$ in time $t_0 \lesssim N^{-1/2}$, for arbitrarily large $N$. An analogous result holds in $B^{3/2}_{2,q}$, $1<q\le\infty$, with the norm inflating by $\ln N$ in time $1/\ln N$ as $N \to \infty$ (Chen et al., 15 Oct 2025).

These constructions rely on initial data with high-frequency oscillations engineered to channel quadratic or cubic nonlinear interactions so as to transfer energy instantaneously to low frequencies; the ill-posedness is 'global' in the sense that it persists for all (even arbitrarily smooth) bounded data in these spaces.

2. Nonuniqueness for Bounded Weak Solutions in Compressible Euler Systems

Beyond norm inflation, global ill-posedness is characterized by wild nonuniqueness: PDEs like the isentropic compressible Euler system in $n \ge 2$ dimensions and certain pressure laws admit infinitely many global-in-time admissible weak solutions for a dense set of energy-bounded initial data.

For pressure $p(\rho) = \rho^\gamma$ with $1<\gamma\le 1+2/n$, and any initial datum in a dense subset $\mathcal D$ of the energy space $L^\gamma \times L^{2\gamma/(\gamma+1)}$, there exist infinitely many admissible (energy-dissipative) global weak solutions. The convex integration framework exploits the flexibility in solving the Euler–Reynolds system with positive 'Reynolds stress' $R$ by superimposing oscillatory corrections that eliminate any smooth defect. Energy-compatible subsolutions are constructed—often via vanishing-viscosity limits—ensuring that the set $\mathcal D$ is dense (Chen et al., 2021).

In 2D, the Chiodaroli–De Lellis–Kreml construction provides explicit Riemann and Lipschitz initial data, bounded away from vacuum, for which uniqueness fails completely. These examples reduce the Cauchy problem for the isentropic Euler system to one solvable by convex integration and demonstrate the absence of any continuous data-to-solution mapping, even for smooth and bounded initial data (Chiodaroli et al., 2013).

3. Mechanisms and Analytical Techniques Enabling Ill-posedness

The onset of global ill-posedness with bounded initial data typically exploits at least one of the following:

• Resonant high-to-low energy transfer: Carefully designed high-frequency initial data interact through nonlinear terms (e.g., quadratic or cubic nonlocalities) to transfer energy rapidly to low frequencies, precluding norm conservation or control (Chen et al., 15 Oct 2025).
• Convex integration schemes: For hyperbolic conservation laws, the lack of convexity or strong entropy conditions generates nonuniqueness via oscillatory solutions that saturate admissibility inequalities and exploit the large convex hull of possible states (Chen et al., 2021, Chiodaroli et al., 2013).
• Criticality of function space embeddings: Underlying many ill-posedness results is the failure of endpoint Sobolev or Besov embeddings, where the Riesz or Calderón–Zygmund operators cease to be bounded below $L^\infty$ or at threshold spaces, enabling instability in critical or subcritical topologies (Bianchini et al., 2023).

The explicit Duhamel–Lagrangian formulation and Littlewood–Paley theory facilitate fine control of nonlocal and nonlinear effects, while commutator and paraproduct estimates provide analytic tools for bounding the troublesome interactions.

4. Comparison with Well-posed Regimes and Conservation Laws

A sharp contrast emerges between ill-posed and well-posed regimes, typically separated by critical exponents or special parameter values:

• In the FOCH model, when $b\le1$, conservation laws yield a priori bounds (e.g., conservation of an $L^{1/b}$-norm of a modified momentum density), forcing global existence and continuous dependence in strong topologies (Chen et al., 15 Oct 2025).
• For 1D isentropic Euler with convex pressure, classical BV theory (Lax–Oleĭnik) ensures unique entropy solutions—even for discontinuous data. This stability collapses in higher dimensions, where geometric flexibility and the lack of BV compactness enable the convex integration machinery to create pathological solutions (Chiodaroli et al., 2013).
• In critical spaces such as $W^{1,\infty}$ for the Boussinesq equation, norm inflation precludes any continuity of the solution map, but subcritical/topologies (e.g., $C^\alpha$, $p<\infty$) retain some well-posedness (Bianchini et al., 2023).

Hence, global ill-posedness with bounded initial data typically coincides with—or signals—the collapse of natural PDE conservation or regularization mechanisms at a sharp threshold.

5. Other Models and Extensions

Norm inflation and breakdown of continuous dependence under bounded initial data have also been established for a range of dispersive and wave-like equations:

• For the generalized improved Boussinesq (gIBq) equation, norm inflation with infinite loss of regularity is proven for initial data in $\langle \nabla \rangle^{-s}(L^2\cap L^\infty)$ for any $s<0$, as well as in Fourier–Lebesgue, modulation, and Wiener amalgam spaces (Roubin, 2023). Explicit high-frequency bump constructions and Duhamel expansions quantify the breakdown of the solution map regardless of data smoothness.
• For the 2D Boussinesq system in vorticity form, as well as the 3D axisymmetric Euler system with swirl, explicit constructions achieve instantaneous norm inflation in $W^{1,\infty}$: bounded, smooth initial data leads to unbounded growth in critical derivative norms in arbitrarily short time, even as other quantities remain bounded (Bianchini et al., 2023).

These results suggest that the criticality of the chosen topology and the nature of nonlinear resonances are central ingredients enabling global ill-posedness for bounded data across diverse PDE models.

6. Implications and Context

The existence of global ill-posedness with bounded initial datum demonstrates the limitations of deterministic PDE theory at or below natural critical thresholds and has significant implications:

• Loss of predictivity: The Cauchy problem becomes non-deterministic in the sense that either multiple solutions exist or the data-to-solution mapping is not continuous, precluding any kind of stable numerical or analytic solution scheme.
• Limits of energy and entropy methods: Even for initial data with uniform bounds in natural (energy or critical Besov/Sobolev) norms, neither uniqueness nor continuous dependence is guaranteed; additional structural or regularity assumptions become indispensable.
• Guidance for model selection and analysis: The parameter regimes (e.g., $b=5/3$ for FOCH) where ill-posedness appears demarcate sharp mathematical transitions between deterministic and pathological behavior, providing a rationale for focusing analytical and numerical effort on well-posed domains.

A plausible implication is that, in numerous nonlinear dispersive or transport equations, any attempt to formulate a well-posed theory must avoid both the critical topology and the nonlinearity-induced resonant transfer mechanisms uncovered by these constructions.

---

References:

• Fifth-order Camassa–Holm norm inflation: (Chen et al., 15 Oct 2025)
• Convex integration and nonuniqueness in compressible Euler: (Chen et al., 2021, Chiodaroli et al., 2013)
• Strong $W^{1,\infty}$ ill-posedness for Boussinesq and 3D axisymmetric Euler: (Bianchini et al., 2023)
• Generalized Boussinesq norm inflation: (Roubin, 2023)