Dissipation and high disorder

Le Chen, M. Cranston, Davar Khoshnevisan, Kunwoo KimarXiv 2014 · DOI · arXiv

This paper, titled "Dissipation and high disorder," explores the long-term behavior of mass in systems governed by stochastic partial differential equations (SPDEs) and their discrete counterparts. It specifically investigates when "noise" in a system is strong enough to cause the total mass to eventually vanish—a phenomenon the authors call global dissipation.

TL;DR

In low dimensions ($d=1, 2$), any amount of environmental noise eventually causes the total mass of a system to dissipate to zero. In higher dimensions ($d \ge 3$), a "phase transition" occurs: the mass survives if the noise is low but dissipates if the noise exceeds a critical threshold.

What Problem Does This Paper Address?

The authors study the Parabolic Anderson Model (PAM), a mathematical framework used to model physical processes like population dynamics in a random environment or heat flow with random sources/sinks.

The central question is about extinction vs. survival. While previous research established that the solution at any specific point ($u_t(x)$) eventually goes to zero (local dissipation), this paper asks if the total mass across the entire infinite space ($\sum u_t(x)$) also vanishes. This is a much harder condition to meet because mass can theoretically "hide" or concentrate in rare, high-growth regions of the environment. The authors aim to determine how the spatial dimension ($d$) and the intensity of the noise ($\lambda$) dictate whether the total mass survives or dissipates.

Methods

The researchers employ a mix of stochastic analysis and probability theory:

  • Lattice Stochastic ODEs: They model the system on a $d$-dimensional integer lattice ($\mathbf{Z}^d$), where mass moves via a random walk (governed by a generator $\mathcal{G}$) and grows/shrinks based on independent Brownian motions at each site.
  • Martingale Theory: They treat the total mass process as a non-negative martingale. By the Martingale Convergence Theorem, the total mass must converge to a limit; the goal is to find when that limit is zero.
  • Fractional Moments: A key technical innovation involves analyzing the "fractional moments" of the mass ($E[m_t^\eta]$ for $\eta < 1$). They derive differential inequalities to show that these moments decay over time.
  • Comparison Theorems: They use mathematical comparison tools to show that increasing the noise parameter $\lambda$ makes dissipation more likely, allowing them to prove the existence of a unique critical threshold ($\lambda_c$).

Datasets

Not discussed in the paper. This is a theoretical mathematics paper; the results are proven through rigorous logical deduction and analytical proofs rather than empirical data or simulations.

Key Insights and Ideas

  • Recurrence vs. Transience: The "aha moment" is the link between the dimension of the space and the behavior of the mass. In 1D and 2D, random walks are "recurrent" (they keep coming back to the same spots), which allows noise to eventually "catch" and dissipate the mass everywhere. In 3D and above, the walk is "transient," allowing mass to potentially escape the dampening effects of noise if the noise is weak enough.
  • Noise as a Sink: Even though the noise term has a mean of zero, the non-linear interaction ($\sigma(u)$) effectively acts as a sink for the total mass over long periods.
  • Disproving Conjectures: The paper provides a rigorous proof for a phase transition in $d \ge 3$, disproving earlier suggestions in the field (notably by Shiga) that extinction might be the universal outcome regardless of dimension.
  • Robustness Across Models: The authors show that their findings for discrete lattices also apply to continuous space (the Stochastic Heat Equation on the real line).

Key Results

  • Dimensions 1 and 2: Global dissipation is inevitable. For any noise level $\lambda > 0$, the total mass $m_t \to 0$ as $t \to \infty$ with probability 1.
  • Dimensions 3 and Higher: A phase transition exists. There is a critical value $\lambda_c$.
  • If $\lambda > \lambda_c$ (High Disorder), the mass dissipates to zero.
  • If $\lambda < \lambda_c$ (Low Disorder), the mass survives with positive probability.
  • Quantitative Decay Rates: For $d=1$, the mass decays at a rate of roughly $\exp(-t^{1/3})$. For $d=2$, it decays much more slowly, at a rate of $\exp(-\sqrt{\log t})$.
  • Lower Bound: They established a universal lower bound for the probability of survival, showing that mass doesn't disappear "too fast" even when it is dissipating.

Limitations and Future Work

  • The Critical Point: The authors state that the behavior at the exact threshold ($\lambda = \lambda_c$) remains an open question. It is unknown if the system dissipates or survives at this precise boundary.
  • Sharpness of Bounds: While they provide decay rates for $d=1$ and $d=2$, they do not know if these specific mathematical bounds are "sharp" (the best possible descriptions of the decay).
  • Generality of $\sigma$: The results assume certain "nice" properties for the nonlinearity function $\sigma$. Exploring more "pathological" or highly irregular functions could be a future direction.

Summary updated

Abstract

Given a field $\{B(x)\}_{x\in\mathbf{Z}^d}$ of independent standard Brownian motions, indexed by $\mathbf{Z}^d$, the generator of a suitable Markov process on $\mathbf{Z}^d,\,\,\mathcal{G},$ and sufficiently nice function $σ:[0,\infty)\to[0,\infty),$ we consider the influence of the parameter $λ$ on the behavior of the system, \begin{align*} \rm{d} u_t(x) = &amp; (\mathcal{G}u_t)(x)\,\rm{d} t + λσ(u_t(x))\rm{d} B_t(x) \qquad[t&gt;0,\ x\in\mathbf{Z}^d], &amp;u_0(x)=c_0δ_0(x). \end{align*} We show that for any $λ&gt;0$ in dimensions one and two the total mass $\sum_{x\in\mathbf{Z}^d}u_t(x)\to 0$ as $t\to\infty$ while for dimensions greater than two there is a phase transition point $λ_c\in(0,\infty)$ such that for $λ&gt;λ_c,\, \sum_{\mathbf{Z}^d}u_t(x)\to 0$ as $t\to\infty$ while for $λ
BibTeX
@misc{chen2014dissipation,
  author = {Le Chen and M. Cranston and Davar Khoshnevisan and Kunwoo Kim},
  title = {{Dissipation and high disorder}},
  year = {2014},
  eprint = {1411.6607},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1411.6607},
}
Figure