Numerical study of probabilistic well-posedness of one dimensional fractional nonlinear wave equations

This paper presents numerical simulations of the one-dimensional fractional cubic defocusing wave equation in a periodic setting, demonstrating that both norm inflation and probabilistic well-posedness can be observed in energy subcritical and supercritical regimes, thereby providing the first numerical evidence of these fine behaviors previously known only theoretically.

The Gist

Imagine you are trying to predict the weather. In a perfect world, if you know the current temperature and wind speed exactly, you can predict the future perfectly. But in the real world, measurements are never perfect. There's always a tiny bit of fuzziness or "noise."

This paper is about a specific type of mathematical weather model: a wave equation. Think of it as a simulation of how ripples move across a pond, but instead of water, it's a mathematical wave that can get very wild and chaotic.

The authors, Wandrille Ruffenach and Nikolay Tzvetkov, are asking a tricky question: "If our starting data is a little bit fuzzy (low regularity), can we still trust the simulation?"

Here is the breakdown of their findings using simple analogies:

1. The Problem: The "Butterfly Effect" Gone Wild

In the world of these equations, there is a known danger called ill-posedness.

• The Analogy: Imagine you are trying to balance a pencil on its tip. If you have a perfect, smooth table (high-quality data), the pencil stays balanced for a while. But if the table is slightly bumpy (low-quality data), the pencil might fall over instantly.
• The Math: For these specific waves, if the starting data is "rough" or "fuzzy," the math says the solution should explode into infinity immediately. It's like the simulation crashing because the input was too messy. This is called Norm Inflation.

2. The Twist: The "Gaussian" Magic Trick

The authors discovered something fascinating. While the math says "chaos!" for rough data, it turns out that if the roughness comes from a specific type of randomness (called Gaussian noise, like static on an old TV), the chaos disappears!

• The Analogy: Imagine you are trying to walk on a tightrope. If you step randomly and wildly, you fall. But if your steps are random in a very specific, natural way (like how leaves fall in the wind), you might actually find a way to stay balanced.
• The Finding: When they used this specific "natural randomness" and chopped it up into manageable pieces (a process called Fourier truncation), the simulation didn't crash. It stayed stable. This is called Probabilistic Well-Posedness. It means: "If you pick a random starting point from this specific bag of possibilities, the simulation will work 99.9% of the time."

3. The Trap: The "Pathological" Approximation

However, there is a catch. The way you approximate that random data matters immensely.

• The Analogy: Imagine you have a blurry photo of a face.
  • Method A (Good): You smooth out the pixels. The face looks a bit fuzzy but recognizable. The simulation works.
  • Method B (Bad): You take that same blurry photo and zoom in on a single, tiny speck of dust, making it huge. The face looks nothing like the original. The simulation explodes.
• The Finding: The authors showed that if you approximate the random data using a "pathological" method (focusing on a weird, concentrated spike), the simulation does explode, even if the starting data looks almost the same. This proves that the stability isn't just about the data; it's about how you prepare the data.

4. The Experiment: Sub-critical vs. Super-critical

The team ran these simulations in two different "universes":

• Energy Sub-critical (The Calm Ocean): Here, the wave's natural spreading (dispersion) is stronger than the chaos of the wave crashing into itself. It's easier to control.
• Energy Super-critical (The Stormy Sea): Here, the chaos is stronger than the spreading. It's much harder to control.
• The Result: Surprisingly, their computer simulations showed that the "Gaussian Magic Trick" (Probabilistic Well-Posedness) worked in both the calm ocean and the stormy sea. Even when the math says it should be impossible, the specific type of randomness saved the day.

5. The Safety Check: The "Fail-Safe"

To make sure their computer code wasn't just hallucinating, they tested a scenario where the math guarantees stability (high-quality data).

• The Analogy: Before testing a new, weird bridge design, you first drive a heavy truck over a standard, known-safe bridge. If the standard bridge holds, your truck and your measuring tools are working correctly.
• The Result: When they used high-quality data, the simulation worked perfectly, and both "Method A" and "Method B" gave the same result. This proved their computer code was reliable and the weird results they saw earlier were real mathematical phenomena, not just computer glitches.

Summary

This paper is a detective story about waves.

1. The Mystery: Rough data usually breaks wave simulations.
2. The Clue: But if the roughness is "random noise" in a specific way, the simulation might actually work.
3. The Villain: If you prepare that noise the wrong way (concentrating it on a single point), the simulation breaks again.
4. The Verdict: Using a computer, they proved that for these specific waves, randomness can actually save the day, keeping the simulation stable even in the most chaotic, "super-critical" environments.

They essentially showed that while the math is fragile, nature's randomness is surprisingly robust, provided you handle it with the right tools.

Technical Summary

1. Problem Statement and Context

The paper investigates the Cauchy problem for the one-dimensional fractional cubic defocusing nonlinear wave equation (fNLW) on a torus $\mathbb{T}$:
$$\partial_t^2 u + D_x^{2\beta} u + u^3 = 0$$
where $D_x = \sqrt{-\Delta}$ is the Fourier multiplier by $|2\pi k|$, and $\beta > 0$ controls the dispersion. The initial data $(u_0, \dot{u}_0)$ belongs to Sobolev spaces $H^s \times H^{s-\beta}$.

Key Theoretical Background:

• Deterministic Ill-posedness: For low regularity initial data ($s < s_c$, where $s_c = 1/2 - \beta$ is the critical regularity), the equation is generally ill-posed. Specifically, there exist sequences of smooth initial data converging to a limit in $H^s$ such that the corresponding solutions diverge in $H^s$ (norm inflation).
• Probabilistic Well-posedness: In contrast, if the initial data is randomized (specifically, fractional Gaussian fields) and approximated via Fourier truncation, well-posedness can be recovered globally in time for certain regimes, even below the critical regularity.
• Pathological Approximation: If the same random initial data is approximated by adding a specific "pathological" perturbation (concentrating at a point while vanishing in norm), norm inflation occurs, and the solution sequence diverges.

The Gap: While these phenomena are established theoretically (mostly in 3D), they have not been numerically observed. The authors aim to simulate these fine behaviors in 1D to verify if probabilistic well-posedness and norm inflation can be distinguished numerically in both energy sub-critical ($\beta > 1/4$) and energy super-critical ($\beta < 1/4$) regimes.

2. Methodology

The authors employ a high-resolution numerical approach using a filtered trigonometric integrator combined with a pseudo-spectral method.

• Spatial Discretization: The domain is discretized using $M$ collocation points (up to $M=2^{25}$). A pseudo-spectral method is used with Fast Fourier Transforms (FFT). To handle the cubic nonlinearity, full dealiasing is performed via zero-padding on a grid of size $2M$.
• Time Integration: A symplectic trigonometric integrator with filtered non-linearity is used to preserve the Hamiltonian structure. The time step $\tau_N$ is adaptive, scaling with the truncation level $N$ to resolve both dispersive and nonlinear effects as regularity decreases.
• Initial Data Approximations: The study compares two distinct approximations of the same random Gaussian initial data $(u_0^\omega, \dot{u}_0^\omega)$:
  1. Fourier Truncation (Probabilistic Case): $u_{0,N}^\omega$ is the standard truncation of the Fourier series up to mode $N$.
  2. Pathological Approximation (Norm Inflation Case): $w_{0,N}^\omega = u_{0,N}^\omega + p_N^s$, where $p_N^s$ is a perturbation that vanishes in $H^s$ but concentrates at a point, designed to trigger ill-posedness.
• Observables:
  • Sobolev Norms: $S_{N, \alpha, \beta}(t)$ and $N_{N, \alpha, \beta}(t)$ track the evolution of the solution in $H^s$.
  • Convergence Metrics: The difference $\Delta S_N$ (or $\Delta N_N$) between solutions at truncation levels $N$ and $N/2$ is computed to test for Cauchy convergence.
  • Hamiltonian Conservation: The relative error of the discretized Hamiltonian is monitored to ensure numerical stability and accuracy.

3. Key Contributions

1. First Numerical Observation of Probabilistic Well-Posedness: The paper provides the first numerical evidence that for low-regularity random initial data, Fourier truncation leads to a convergent sequence of solutions in both energy sub-critical and super-critical regimes.
2. Numerical Demonstration of Norm Inflation: The authors successfully simulate the "pathological" scenario where a vanishing perturbation to the initial data causes the solution norm to blow up instantaneously as $N \to \infty$, confirming the theoretical ill-posedness.
3. Regime Comparison: The study systematically compares energy sub-critical ($\beta = 1/3$) and energy super-critical ($\beta = 1/8$) cases, showing that while probabilistic well-posedness holds in both, norm inflation is more pronounced in the super-critical regime due to the dominance of nonlinearity over dispersion.
4. Deterministic "Fail-Safe" Verification: The authors verify that in the deterministic well-posed regime (high regularity), both approximation methods yield the same convergent solution, validating the robustness of their numerical scheme.

4. Key Results

• Probabilistic Well-Posedness (Sections 2.4 & 2.5):

  • For low regularity ($\alpha - 1/2 < 1/2 - \beta$) and Fourier truncation, the Sobolev norms $S_{N, \alpha, \beta}(t)$ remain bounded uniformly in time as $N \to \infty$.
  • The difference between solutions at different truncation levels ($\Delta S_N$) converges to zero, indicating the sequence is Cauchy in $C([0, T], H^s)$.
  • This behavior holds for both $\beta = 1/3$ (sub-critical) and $\beta = 1/8$ (super-critical).

• Norm Inflation (Section 2.5):

  • When using the pathological approximation (adding $p_N^s$), the Sobolev norms $N_{N, \alpha, \beta}(t)$ diverge as $N \to \infty$.
  • The divergence is instantaneous in the limit, with the norm growing unbounded for any $T > 0$.
  • The divergence is significantly faster in the super-critical case ($\beta = 1/8$) compared to the sub-critical case, consistent with the heuristic that nonlinearity dominates dispersion in the super-critical regime.
  • The difference $\Delta N_N$ fails to converge to zero, confirming the lack of convergence in the solution space.

• Deterministic Well-Posedness (Section 2.6):

  • In the high regularity regime ($s > 1/2 - \beta$), both the Fourier truncation and the pathological approximation converge to the same unique solution.
  • This serves as a "sanity check," proving that the observed divergence in the low-regularity pathological case is a physical phenomenon of the PDE, not a numerical artifact.

• Hamiltonian Conservation (Section 2.7):

  • The relative error in the discretized Hamiltonian remains small ($< 10^{-7}$) and does not grow with $N$, even for the pathological cases. This confirms that the observed norm inflation is not due to numerical instability but is intrinsic to the equation's dynamics.

5. Significance

This work bridges the gap between rigorous PDE theory and numerical simulation for stochastic nonlinear wave equations. Its significance lies in:

• Validation of Theory: It numerically confirms that the choice of approximation for random initial data is critical: Fourier truncation yields well-posedness, while pathological perturbations yield ill-posedness.
• Super-critical Insights: It demonstrates that probabilistic well-posedness is not limited to energy sub-critical regimes; it persists in the energy super-critical regime for 1D fractional equations, a regime where global existence is otherwise an open problem.
• Methodological Benchmark: The study establishes a robust numerical framework (filtered trigonometric integrators with adaptive steps) capable of resolving the delicate interplay between dispersion, nonlinearity, and low regularity, setting a standard for future investigations into blow-up phenomena and statistical properties of nonlinear wave equations.