🔢 mathematics

Long-time behaviour of dynamical systems driven by bounded mixing noises

This paper establishes exponential mixing for both finite- and infinite-dimensional dissipative dynamical systems driven by bounded mixing noises by introducing a generalized class of random forcings, lifting the dynamics to a history space, and applying a Doeblin coupling argument via the Kantorovich functional.

Original authors: Peng Gao, Sergei Kuksin

Published 2026-07-08✓ Author reviewed ⓘ

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Peng Gao, Sergei Kuksin

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to predict the weather, or the movement of a crowd, or the flow of water in a river. These are dynamical systems. They change over time based on their current state and some outside forces.

Now, imagine those outside forces are a bit chaotic—like random gusts of wind or unpredictable pushes. In math, we call these "random forces." Usually, if you push a system randomly enough, it eventually forgets where it started and settles into a predictable, steady pattern of behavior. This is called mixing.

This paper, written by Peng Gao and Sergei Kuksin, is a guidebook on how to prove that certain complex systems will indeed "forget" their past and settle down, even when the random forces pushing them are bounded (they don't go off to infinity) and mixing (they have their own internal randomness).

Here is the breakdown of their work using simple analogies:

1. The Two Types of Systems

The authors look at two kinds of systems:

Discrete-time: Like taking a photo of a bouncing ball every second. You see where it is at second 1, second 2, second 3, etc.
Continuous-time: Like watching a video of the ball moving smoothly.

The paper's clever trick is to say: "Let's just treat the video as a series of photos." They take the continuous video, chop it into one-second chunks, and turn it into a discrete photo series. This allows them to do all their heavy math on the simpler "photo" version, and then apply the results back to the "video" version.

2. The "Memory" of the Random Force

The random forces (let's call them the "wind") aren't just pure chaos. The authors assume the wind has a specific personality:

It has a short memory: If you know the wind's history for the last few seconds, you can predict its next move with decent accuracy. It doesn't hold grudges from the distant past.
It visits the "center": The wind occasionally calms down and blows gently near zero. It doesn't just scream at high speeds forever.
It's smooth: The wind doesn't jump erratically; it flows in a way that allows for mathematical prediction.

3. The "Control" Problem

The system they are studying (the ball, the fluid, the crowd) needs to be controllable.

The Analogy: Imagine a boat in a river. If the river current (the random force) can push the boat in any direction you want, the boat can eventually reach any spot in the river.
The Math: The authors prove that if the random force can "push" the system in enough different directions (a condition they call linearised controllability), the system will mix.

4. The Magic Tool: "Coupling"

How do they prove the system forgets its past? They use a technique called Coupling.

The Analogy: Imagine two identical boats starting at different places in the river.
- Boat A starts at the North bank.
- Boat B starts at the South bank.
- Normally, they might drift apart. But the authors invent a "magic wind" that is the same for both boats, but slightly adjusted.
- They show that if you steer both boats with this specific "magic wind," they will eventually crash into each other (or get so close they are indistinguishable).
The Result: Once the two boats crash together, they stay together. This means it doesn't matter where you started; the final destination is the same. This proves the system has "mixed."

5. Finite vs. Infinite Dimensions

Finite Dimensions (The Simple Case): This is like a ball moving in a 3D room. The authors prove that for these systems, the "forgetting" happens exponentially fast. The difference between two starting points shrinks by a fixed percentage every second.
Infinite Dimensions (The Complex Case): This is like trying to track the movement of every single molecule in a cloud of gas. This is much harder because there are infinite variables.
- The authors had to tweak their "magic wind" and their "control" rules.
- Instead of proving the boats crash perfectly (which is impossible in infinite dimensions), they prove the boats get so close that any reasonable observer can't tell them apart. They call this the dual-Lipschitz distance.

6. Real-World Applications Mentioned

The paper doesn't just stay in theory; they show how this applies to real equations:

Ordinary Differential Equations (ODEs): These describe things like a pendulum or a population of animals. The authors show that if you shake these systems with the right kind of random noise, they will settle into a stable pattern quickly.
Primitive Equations of Atmospheric Dynamics: This is the big one. These are the equations used to model the ocean and the atmosphere (weather).
- The authors apply their theory to these massive, complex fluid equations.
- They prove that even though the atmosphere is huge and complex, if you add a bounded, random "wind" to it, the system will eventually mix and forget its initial weather conditions.

Summary

In short, this paper is a mathematical proof that chaos leads to order. If you have a complex system (like the weather) and you shake it with the right kind of random, bounded force, the system will eventually stop caring about where it started and settle into a predictable, steady rhythm. The authors provided a new, more flexible toolkit to prove this, especially for the massive, infinite systems like the Earth's atmosphere.

Technical Summary: Long-time behaviour of dynamical systems driven by bounded mixing noises

Problem Statement
This work investigates the long-time statistical behavior, specifically the mixing properties, of dissipative dynamical systems driven by bounded random forces. The study addresses both discrete-time systems of the form $u_k = S(u_{k-1}, \eta_k)$ and continuous-time systems $\dot{u}(t) = F(u(t), \eta(t))$ . The primary objective is to establish exponential mixing in the total variation metric (for finite-dimensional phase spaces) and the dual-Lipschitz metric (for infinite-dimensional spaces) under conditions where the random forcing $\eta$ is bounded and satisfies specific mixing and regularity properties. The work extends the framework developed by Kuksin and Shirikyan (2025) to broader classes of non-stationary processes and continuous-time systems with continuous sample paths.

Methodology
The authors employ a "Markov lifting" strategy combined with a coupling argument based on the Kantorovich functional. The core methodological steps include:

Reduction to Discrete Time: Continuous-time systems are reduced to discrete-time random dynamical systems (RDS) via time-one maps. For continuous paths, the noise over an interval $I_k = [k-1, k)$ is treated as an element of a Hilbert space of curves.
Markovian Lifting: The original non-Markovian system (where the noise may have memory) is lifted to a Markov process on an extended state space $\mathcal{X} = X \times K$ , where $X$ is the phase space and $K$ is the space of noise histories. The dynamics on $\mathcal{X}$ are defined by a map $\mathcal{S}(U, \eta) = (S(v, \eta), (\xi, \eta))$ .
Doeblin Coupling via Kantorovich Functional: To prove exponential mixing, the authors construct a coupling of two trajectories starting from different initial conditions. This involves:
- Local Stabilization: Using linearized controllability assumptions to construct a map $\Phi$ that "squeezes" the distance between two states when they are close.
- Recurrence: Utilizing a "recurrence to the origin" condition to ensure trajectories enter a small neighborhood of the origin with positive probability.
- Iterative Decay: Applying the Kantorovich functional to show that the distance between the laws of the coupled trajectories decays exponentially.
Handling Infinite Dimensions: For infinite-dimensional phase spaces, the standard linearized controllability assumption is replaced by an "Approximate Linearized Controllability" (ALC) condition involving a finite-dimensional determining subspace and a sequence of finite-dimensional subspaces approximating the noise space.

Key Assumptions
The results rely on the following structural assumptions on the system and the noise:

Dissipativity: The unforced system is strictly dissipative, and the forced system possesses a compact absorbing set.
Regularity (SF): The conditional distributions of the noise with respect to its past are Lipschitz continuous in the total variation distance (analogous to the strong Feller property).
Recurrence (RZ): The noise process has a positive probability of entering and remaining in a small neighborhood of the origin for a fixed duration, regardless of the past.
Controllability/Non-degeneracy:
- Finite-dimensional: A linearized controllability condition where the derivative of the map with respect to the noise, restricted to a finite-dimensional subspace, is surjective. The conditional noise distribution on this subspace must admit a Lipschitz density.
- Infinite-dimensional: An approximate version where the image of the noise derivative contains a finite-dimensional determining subspace, and the noise satisfies a decomposability and Lipschitz property (DLP) on a sequence of finite-dimensional subspaces.

Main Results

Finite-Dimensional Case (Theorem A/3.3): Under the stated assumptions, the discrete-time system is exponentially mixing in the total variation distance. There exists a unique stationary process (in distribution) such that any solution converges to it exponentially fast.
Infinite-Dimensional Case (Theorem B/5.2): For systems with infinite-dimensional phase spaces (e.g., PDEs), the system is exponentially mixing in the dual-Lipschitz distance. The total variation mixing is generally not claimed for the infinite-dimensional case due to the lack of absolute continuity of measures in infinite dimensions.
Process Reconstruction (Corollary C): The paper demonstrates that the mixing properties of the lifted system imply that a random process can be uniquely reconstructed from its conditional transition probabilities if those probabilities satisfy the regularity and recurrence conditions.
Applications to ODEs and PDEs:
- ODEs: The results apply to ordinary differential equations driven by bounded random processes with $H^1_{loc}$ trajectories (continuous paths). The authors show how to decompose the noise to satisfy the required hypotheses.
- PDEs: The theory is applied to the randomly perturbed primitive equations of atmospheric dynamics (3D). The authors establish exponential mixing in the dual-Lipschitz metric for these equations driven by bounded continuous random forces.

Significance and Claims
The paper claims to provide a more accessible version of the approach introduced in Kuksin and Shirikyan (2025) while significantly extending its applicability. Specifically:

It removes the requirement for the noise process to be stationary, accommodating non-stationary bounded processes.
It extends the framework to continuous-time systems with continuous sample paths (specifically $H^1_{loc}$ trajectories), which were not covered in the previous work that focused on $L^2_{loc}$ processes.
It relaxes the density assumptions on the noise space in the infinite-dimensional setting, allowing for cases where the union of the finite-dimensional subspaces is not dense in the full noise space.
The methodology successfully handles the primitive equations of atmospheric dynamics, a significant non-linear PDE system, proving their exponential mixing under bounded continuous forcing, a regime where well-posedness is known but mixing properties were previously unestablished for continuous forces.

The authors emphasize that their approach relies on lifting the dynamics to a Markov process on a history space and applying a coupling argument, which allows them to bypass the strong Feller property of the original system, which often fails in infinite dimensions or for non-Markovian noise.