Two approaches to average stochastic perturbations of integrable systems

Basic Information

• Paper ID: 2509.17159
• Title: Two approaches to average stochastic perturbations of integrable systems
• Author: Sergei Kuksin
• Classification: math.DS (Dynamical Systems)
• Publication Date: October 11, 2025 (arXiv v2)
• Paper Link: https://arxiv.org/abs/2509.17159

Abstract

This paper discusses two methods for studying the long-time and infinite-time behavior of solutions to integrable Hamiltonian systems under small stochastic perturbations, and compares these results with those obtained for deterministic perturbations of integrable systems.

Research Background and Motivation

Problem Background

1. Classical Problem: Studying the behavior of small perturbations of integrable Hamiltonian systems on large time scales is a classical problem in dynamical systems, dating back to the late 18th century
2. Traditional Approaches: Existing research primarily focuses on two scenarios:
   • A) Perturbations are small smooth vector fields
   • B) Perturbations are small Hamiltonian fields
3. New Challenge: This paper addresses a third case—stochastic perturbations, which are significant both theoretically and practically

Research Motivation

• Stochastic perturbations better reflect the noise effects in actual physical systems compared to deterministic perturbations
• Existing averaging theory is primarily developed for deterministic systems, with the stochastic case being relatively underdeveloped
• New mathematical tools are needed to handle long-time dynamical behavior under stochastic perturbations

Core Contributions

1. Proposes two distinct approaches to study the averaging problem for stochastic perturbations of integrable systems
2. Establishes a fast-slow system method by transforming the problem into the study of slow variables through action-angle coordinates
3. Develops an effective equation method that directly constructs averaged equations in the original coordinate system
4. Proves convergence theorems providing the limiting behavior of action variables under stochastic perturbations
5. Provides detailed comparisons with deterministic perturbation results, revealing the stabilizing effect of stochasticity

Detailed Methodology

Basic Setup

Consider a Birkhoff integrable Hamiltonian system in the symplectic space $(R^{2n}_{x,y}, dx \wedge dy)$ whose Hamiltonian function depends only on action variables $I_j = (x_j^2 + y_j^2)/2$. Introducing complex coordinates $v_j = x_j + iy_j$, the system can be written as:

$\frac{\partial}{\partial t}v_k(t) = i\nabla_k H(I)v_k, \quad k = 1, \ldots, n$

ε-Small Stochastic Perturbations

Consider the following stochastically perturbed system (slow time $\tau = εt$):

$\dot{v}_k(\tau) = iε^{-1}\nabla_k H(I)v_k + P_k(v) + \sum_{j=1}^{n_1} B_{kj}(v)\dot{\beta}^c_j(\tau)$

where $\{\beta^c_k(t)\}$ are standard independent complex Wiener processes.

Basic Assumptions

(A1) Anosov Non-degeneracy Condition: For almost all $I \in R^n$, the components of the frequency vector $\nabla H(I)$ are linearly independent over the integers

(A2) Local Lipschitz condition and polynomial growth condition

(A3) Well-posedness of the equation and uniform boundedness of moments

(A4) Non-degenerate noise: The rank of matrix $B(v)$ equals $n$

First Approach: Fast-Slow Systems

Action-Angle Coordinate Transformation

Introduce action-angle coordinates $(I, \phi)$:

• $I_j = \frac{1}{2}|v_j|^2$
• $\phi_j = \arg v_j$

Slow Variable Equation

The action variables satisfy: $\dot{I}^ε(\tau) = F(v^ε) + G(v^ε)\dot{\beta}(\tau)$

This is a fast-slow system where the actions $I^ε_k$ are slow variables and the angles $\phi^ε_k$ are fast variables.

Averaged Equation

By averaging over the fast variables, we obtain: $\dot{I}(\tau) = \langle\langle F \rangle\rangle(I) + \langle\langle G \rangle\rangle(I)\dot{\beta}(\tau)$

where averaging is defined as: $\langle\langle F \rangle\rangle(I) = \frac{1}{(2\pi)^n} \int_{T^n} F(I, \phi) d\phi$

Main Results

Theorem 2.1: The action curves $I^ε(\tau)$ of solutions $v^ε(\tau)$ are precompact in distribution, and every limit process is a weak solution of the averaged equation.

Limitations: When the matrix $B(v)$ is not constant, the averaged coefficients may only have Hölder-1/2 smoothness, leading to uniqueness issues for solutions.

Second Approach: Effective Equations

Direct Averaging

Inspired by Krylov-Bogolyubov averaging, we directly construct effective equations in complex coordinates:

$\dot{a}(\tau) = \langle\langle P \rangle\rangle(a) + \langle\langle B \rangle\rangle(a)\dot{\beta}^c(\tau)$

where averaging is defined through rotation operators: $\langle\langle P \rangle\rangle(a) = \frac{1}{(2\pi)^n} \int_{T^n} \Phi_\omega \circ P(\Phi_{-\omega}a) d\omega$

Main Advantages

• Avoids singularities in action-angle coordinates
• Preserves the smoothness of the original equation
• Always admits a unique solution

Core Theorems

Theorem 3.1: Under assumptions A1)-A4), the effective equation has a unique solution, and as $ε \to 0$, $I^ε(\tau)$ converges in distribution to the action of $a(\tau)$.

Theorem 3.3: If additionally assumption A5) (long-time estimates) holds, the convergence is uniform in time.

Theorem 3.4: If $P(v) = P_1(v) + P_2(v)$, where $P_2$ is a Hamiltonian field, the effective equation can be modified to average only the non-Hamiltonian part.

Experimental Setup and Applications

Application Scenarios

5.1 Oscillator Chains: Consider nonlinear oscillator chains: $\ddot{q}_k = -Q(q_k), \quad k = 1, \ldots, n$ where the potential function $Q$ is smooth, convex, and odd, with $Q(q) = αq + βq^3 + o(|q|^4)$ near the origin.

5.2 Damped/Driven Hamiltonian Systems: Consider the equation: $\dot{v}_k(\tau) = iε^{-1}\nabla_k H(I(v))v_k + 2i\frac{\partial}{\partial \bar{v}_k}h(v) - ν_k v_k + b_k \dot{\beta}^c_k(\tau)$

For such systems, the effective equation simplifies to: $\dot{a}_k(\tau) = -ν_k a_k + b_k \dot{\beta}^c_k(\tau)$

This represents $n$ independent complex Ornstein-Uhlenbeck processes, which can be solved explicitly.

Comparison of Main Results

Comparison with Deterministic Perturbations

6.1 Hamiltonian Perturbations: The Nekhoroshev theorem states that under analyticity conditions, action variables remain $ε^b$-close to their initial values over exponentially long times $\tau \lesssim \exp ε^{-a}$.

Contrast: The stochastic perturbation results in this paper allow arbitrary initial data and control the behavior of action variables over arbitrarily long times, benefiting from the stabilizing effects of noise and friction.

6.2 Arbitrary Perturbations: The Neishtadt-Bakhtin theorem requires random parameters to avoid a small measure set, whereas the results here hold for all initial data.

Technical Innovations

1. Avoiding Singularities: The effective equation method avoids singularities of action-angle coordinates near the origin
2. Preserving Smoothness: Averaged coefficients maintain the smoothness of the original equation
3. Unified Framework: The two methods provide complementary perspectives applicable under different technical conditions
4. Long-time Stability: Exploits the mixing properties of stochasticity to achieve long-time control

Conclusions and Discussion

Main Conclusions

1. The effective equation method is more reliable in general cases, avoiding technical difficulties of the fast-slow system approach
2. Stochastic perturbations exhibit stronger stabilizing effects compared to deterministic perturbations
3. Hamiltonian parts of perturbations do not affect the dynamics of action variables after averaging

Limitations

1. Requires global Birkhoff integrability assumption (partially mitigable through local versions)
2. Non-degenerate noise assumption may be too strong for some applications
3. Convergence is in distribution, not almost sure convergence

Future Directions

1. Extension to partially degenerate noise cases
2. Study of more general infinite-dimensional systems (stochastic partial differential equations)
3. Development of finer convergence rate estimates

In-Depth Evaluation

Strengths

1. Theoretical Completeness: Provides two complementary methods with a complete theoretical framework
2. Technical Innovation: The effective equation method cleverly circumvents technical difficulties of traditional approaches
3. Broad Applicability: Results apply to various physical systems, such as oscillator chains and damped systems
4. Thorough Comparison: Detailed comparison with deterministic cases reveals the essential role of stochasticity

Weaknesses

1. Strong Assumptions: Global integrability and non-degenerate noise assumptions limit applicability
2. Convergence Type: Only distributional convergence is proven; almost sure convergence results are absent
3. Numerical Verification: Lacks numerical experiments validating theoretical predictions

Impact

1. Theoretical Contribution: Provides new tools for averaging theory of stochastic dynamical systems
2. Applied Value: Significant implications for non-equilibrium statistical physics
3. Methodological: The effective equation method may inspire research on other stochastic systems

Applicable Scenarios

• Stochastic Hamiltonian systems with multiple time scales
• Stochastic oscillator models in non-equilibrium statistical mechanics
• Integrable partial differential equations under stochastic perturbations

References

Primarily references classical texts by Arnold et al., particularly "Mathematical Aspects of Classical and Celestial Mechanics," and the author's previous important works in this field, especially the series of papers in collaboration with Huang and Piatnitski.