Uniform exponential contraction for viscous Hamilton–Jacobi equations

Konstantin Khanin, Ke Zhang, Lei ZhangCalculus 2023 · DOI · arXiv

Abstract

This research addresses a fundamental gap in the asymptotic analysis of viscous Hamilton–Jacobi (HJ) equations. While it is established that solutions settle into a unique stationary state for both viscous ($\nu > 0$) and inviscid ($\nu = 0$) regimes, the mechanisms driving these processes are traditionally viewed as distinct. This paper proves the existence of a uniform exponential contraction rate $\lambda > 0$ that remains bounded away from zero as viscosity vanishes. By integrating Weak KAM theory with the study of Markov operators, the authors establish a robust link between stochastic diffusion and deterministic hyperbolicity.

Introduction

The viscous Hamilton–Jacobi equation is a cornerstone of mathematical physics, describing systems where evolution is governed by a potential field subject to diffusive noise (viscosity, $\nu$). In periodic systems, the long-term behavior of the solution $\phi^\nu$ is characterized by its convergence to a stationary state $\psi^\nu$.

Historically, the rate of this convergence, $\lambda(\nu)$, has been analyzed through two separate lenses. In the viscous case, contraction is driven by the Markov mechanism—essentially the smoothing effect of diffusion. However, standard estimates for this rate typically vanish as $\nu \to 0$. Conversely, in the inviscid limit, contraction is driven by the dynamical mechanism, specifically the hyperbolicity and "stretching and folding" of action-minimizing orbits. The central challenge addressed here is the "uniformity problem": proving that the contraction rate does not collapse during the transition from stochastic to deterministic dynamics.

Problem Statement

The primary difficulty in establishing a uniform rate lies in the failure of standard diffusive estimates at low viscosity. As $\nu$ approaches zero, the Markov mechanism weakens, yet physical intuition suggests that the underlying dynamical structure of the system should maintain a consistent rate of stabilization. Prior to this work, a rigorous proof was missing to show that a single $\lambda$ could bound the contraction for all $\nu \geq 0$, ensuring that the system does not take infinitely long to stabilize as noise disappears.

Methodology

The authors focus on the discrete-time "kicked forcing" model, where the potential is applied in periodic pulses. Their approach utilizes a multi-disciplinary framework:

  • Hopf-Cole Transformation: The non-linear HJ equation is converted into a linear inhomogeneous heat equation. This allows the evolution of the system to be analyzed through the spectral properties of linear operators.
  • Weak KAM Theory: This theory is employed to analyze the "minimal paths" of the system. In the inviscid limit, the solution is determined by paths that minimize a specific action functional; the authors use this to characterize the geometry of the state space.
  • Markov Operators and Lyapunov Functions: The evolution of the heat equation is treated as a sequence of Markov operators. By applying a Lyapunov function approach—common in ergodic theory—the authors demonstrate that these operators contract the functional space at a consistent rate.
  • Bootstrap Argument: To overcome the limitations of standard asymptotic methods (like Laplace’s method) in high-dimensional settings over long time scales, the authors developed a bootstrap strategy. This method scales short-term uniform estimates into global, long-term contraction bounds.

Key Insights

The Hybrid Mechanism

The core contribution of the paper is the conceptualization of a "hybrid mechanism." The authors demonstrate that the Markov and dynamical mechanisms are not mutually exclusive. Instead, the hyperbolicity of the deterministic system provides a structural backbone that allows the Markov chain to converge uniformly, even when the diffusion coefficient is infinitesimally small.

Conjugate Kernels and Spectral Gaps

By "conjugating" the operators using the stationary solution of the inviscid equation, the authors center their analysis on the global minimizer of the state space. They prove that the "spectral gap"—the distance between the primary and secondary eigenvalues of the evolution operator—does not close as $\nu \to 0$. This ensures that the partition function of the system remains well-behaved in the vanishing viscosity limit.

Results

The research culminates in Theorem A, which states that for a generic periodic potential $F$, there exists a uniform constant $\lambda > 0$ such that for all sufficiently small $\nu$, the solution $\phi^\nu$ converges to the stationary solution $\psi^\nu$ at a rate of at least $e^{-\lambda n}$.

Furthermore, the paper provides a rigorous link between the trajectories of the viscous "random walk" and the "minimizing paths" of the Lagrangian. It shows that as $\nu \to 0$, the distribution of the viscous trajectories converges to the deterministic minimizing path, providing a complete bridge between the two regimes.

Discussion and Future Work

While this work provides a definitive answer for the discrete-time case, several avenues for extension remain. The authors expect these results to hold for continuous-time systems, though the technical implementation requires further detail. Additionally, the current results assume a generic potential with a single global minimum. Future research may explore potentials with multiple minimizers, where "metastability" could complicate the uniform contraction rate. Finally, the authors intend to apply this methodology to systems with random forcing, where the "kicks" themselves are stochastic variables.

Summary updated

BibTeX
@inproceedings{khanin2023uniform,
  author = {Konstantin Khanin and Ke Zhang and Lei Zhang},
  title = {{Uniform exponential contraction for viscous Hamilton–Jacobi equations}},
  booktitle = {Calculus},
  year = {2023},
  doi = {10.1007/s00526-022-02423-0},
}
Figure