Noise sensitivity for stochastic heat and Schrödinger equation

Yu Gu, Tomasz Komorowski

TL;DR

This paper explores how solutions to the stochastic heat and Schrödinger equations become increasingly "chaotic" over time. The authors prove that the order of noise interaction (Wiener chaos) that dominates the solution follows a Gaussian distribution for large times and demonstrate that even a tiny perturbation in the underlying noise—on the scale of $1/t$—is enough to completely decorrelate the system.

What Problem Does This Paper Address?

In the study of Stochastic Partial Differential Equations (SPDEs), we often want to understand how a system's state depends on the random noise driving it. This is typically analyzed using Wiener chaos expansion, which decomposes a random variable into a sum of terms representing increasingly complex interactions with the noise (linear, quadratic, etc.).

The "Fourier spectrum" of a solution at time $t$ tells us the probability distribution of these chaos orders. A key question in the field of noise sensitivity is: how much do we need to change the underlying noise to make the new solution look completely different from the original? This paper specifically tackles the asymptotic behavior of this spectrum for two fundamental equations—the stochastic heat equation (SHE) and the Schrödinger equation—as time $t$ goes to infinity.

Methods

The authors employ several sophisticated tools from stochastic analysis and Malliavin calculus:

  • Wiener Chaos Expansion: They represent the solutions as infinite series of iterated Itô integrals.
  • Ornstein-Uhlenbeck Semigroups: To study noise sensitivity, they use these semigroups to define a "perturbed" version of the noise and calculate the correlation between the original and perturbed solutions.
  • Feynman-Kac Formula: For the stochastic heat equation, they use this representation to express the solution as an expectation over Brownian motion paths, which simplifies the calculation of moments.
  • Analytic Continuation: In the proof for the heat equation, the authors use analytic continuation to move from Laplace transforms (which are easier to compute) to characteristic functions to prove convergence in law.

Datasets

Not discussed in the paper. This is a purely theoretical mathematical study; no empirical datasets or benchmarks were used.

Key Insights and Ideas

  • The $1/t$ Scaling Law: The most significant insight is that the "onset of chaos" occurs on the scale of $1/t$. This means as time progresses, the system becomes exponentially more sensitive to noise; a perturbation that was negligible at $t=1$ becomes disruptive at $t=100$.
  • Poisson-like Behavior: The authors find that the random variable $N_t$ (representing the chaos order) behaves similarly to a Poisson distribution with an intensity parameter that grows linearly with $t$.
  • Universal Gaussian Limit: Despite the different physical natures of the heat equation (diffusion) and the Schrödinger equation (wave-like/quantum), both exhibit the same asymptotic behavior: after centering and rescaling, their Fourier spectra converge to a standard Gaussian distribution.
  • Analogy to Geometric Brownian Motion: The paper points out that the mechanism driving this chaotic behavior is analogous to the way geometric Brownian motion evolves, providing a helpful mental model for researchers in the field.

Key Results

  • Theorem 1.1 (Central Limit Theorem): The main result proves that for both equations, the rescaled chaos order $N_t$ satisfies: $$\frac{N_t - \mu t}{\sigma \sqrt{t}} \Rightarrow N(0, 1) \text{ as } t \to \infty$$ where $\mu$ and $\sigma$ are constants depending on the specific equation.

  • Decorrelation Threshold: The authors show that if the noise perturbation strength $s$ scales as $t^{-\alpha}$:

  • If $\alpha > 1$ (very small perturbation), the solutions remain perfectly correlated (correlation $\to 1$).
  • If $\alpha < 1$ (larger perturbation), the solutions become completely decorrelated (correlation $\to 0$).
  • This identifies $s \sim t^{-1}$ as the critical threshold for noise sensitivity.

Limitations and Future Work

  • Linearity: The study is restricted to linear SPDEs with multiplicative noise. While these are fundamental, many real-world systems are nonlinear.
  • Nonlinear SPDEs (KPZ Equation): The authors acknowledge that proving similar results for nonlinear equations, such as the Kardar-Parisi-Zhang (KPZ) equation or the free energy of directed polymers, is a much harder, open problem.
  • Conjecture for Polymers: They suggest a "plausible conjecture" that for directed polymers, the chaos order might scale as $t^{1/3}$ rather than $t$, reflecting the different universality class (KPZ vs. Gaussian). Proving this remains a significant challenge for the community.

Summary updated

Abstract

In this note, we consider the stochastic heat and Schrödinger equation, and show that, at time $t$, the onset of the chaos occurs on the scale of $1/t$, and the Fourier spectrum of the solution is asymptotically Gaussian after centering and rescaling.
BibTeX
@article{gu2025noise,
  author = {Yu Gu and Tomasz Komorowski},
  title = {{Noise sensitivity for stochastic heat and Schrödinger equation}},
  journal = {Bulletin},
  year = {2025},
  doi = {10.4064/ba250422-15-9},
}
Figure