Higher order return times for φ-mixing measures

Basic Information

• Paper ID: 2510.14116
• Title: Higher order return times for φ-mixing measures
• Authors: Nicolai T A Haydn (University of Southern California), Gin Park
• Classification: math.PR (Probability Theory), math.DS (Dynamical Systems)
• Publication Date: October 15, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2510.14116

Abstract

This paper investigates φ-mixing measures and proves that the limiting distribution of higher order return times is compound Poisson when the target set shrinks to a zero-measure set. The methodology employed generalizes the first entrance time approach proposed by Galves and Schmitt in 1997 to higher order cases.

Research Background and Motivation

Problem Background

1. Historical Development: The study of entrance time distributions primarily originated in the past three decades, initially by Pitskel using symbolic dynamics for axiom A maps to study equilibrium states, followed by Hirata using Laplace transforms and spectral methods of transfer operators to obtain similar results.
2. Core Problem: Investigating the statistical laws governing the return of points to specific regions in dynamical systems, particularly the limiting distribution behavior when target sets gradually shrink.
3. Existing Limitations:
   • Early work focused primarily on first-order return times
   • Research on higher-order return times for φ-mixing systems was insufficient
   • More general methods were needed to handle arbitrary zero-measure sets
4. Research Significance:
   • Understanding long-term statistical behavior of dynamical systems
   • Providing theoretical foundations for extreme value theory and rare event analysis
   • Generalizing the classical Kac theorem to more general settings

Core Contributions

1. Theoretical Generalization: Extending the Galves-Schmitt method from first-order entrance times to higher-order return times
2. Mixing Condition Optimization: Demonstrating the advantages of φ-mixing conditions over α-mixing conditions, allowing slower decay rates
3. Compound Poisson Limit: Establishing compound Poisson distribution limit theorems for higher-order return times
4. Unified Framework: Providing a unified method for handling arbitrary zero-measure sets, including single points and more general geometric structures

Methodology Details

Problem Setup

Let (Ω, μ, T) be a measure-preserving dynamical system, and let U_n ⊂ Ω be a nested sequence of positive measure sets whose intersection Γ = ∩_n U_n is a zero-measure set. Define the counting function:

$Z^N_U = \sum_{j=0}^{N-1} \chi_U \circ T^j$

The study examines the limiting distribution of Z^{N_n}_ when μ(U_n) → 0.

Core Theoretical Framework

1. φ-Mixing Condition

A measure μ is left φ-mixing if there exists a decreasing function φ(k) → 0 such that: $|μ(B ∩ T^{-n-k}C) - μ(B)μ(C)| ≤ φ(k)μ(B)$ holds for all B ∈ σ(A^n), C ∈ σ(∪_{ℓ=1}^∞ A^ℓ).

2. Clustering Probability Parameters

Define the clustering probability: $λ_k(L,U) = \frac{μ(Z^L_U = k)}{μ(τ_U ≤ s)}$

where τ_U(x) = inf{j ≥ 1: T^jx ∈ U} is the first hitting time.

3. Main Theorems

Theorem 2.1 (Single Limit Theorem): If μ is φ-mixing, s_n → ∞ and s_n μ(τ_ ≤ s_n) → 0, then: $λ_k = \lim_{n→∞} λ_k(s_n, U_n)$

Theorem 3.1 (Compound Poisson Limit): Let A be a finite generating partition, μ be left φ-mixing with φ(x) = O(x^{-p}), p > 0. If U_n ∈ σ(A^n), μ(∩n U_n) = 0, and there exists a sequence s_n such that P(Z^{s_n} ≥ 1) → 0, s^η_n P(Z^{s_n}_ ≥ 1) → ∞ for some η ∈ (0,1), then:

$Z^{N_n}_{U_n} \xrightarrow{d} W$

where W is a compound Poisson random variable with parameters t and {λ̂_k}, and N_n = ts_n/P(Z^{s_n}_ ≥ 1).

Technical Innovations

1. Generalized Convolution Formula

Lemma 4.1: For 0 < s ≤ t and 0 < Δ < s/2: $\left|μ(Z^{t+s}_U = k) - \sum_{j=0}^k μ(Z^t_U = j)μ(Z^s_U = k-j)\right| < μ(τ^{k+1}_U > t-Δ)(4μ(τ_U ≤ Δ) + 3φ(Δ-n))$

This generalizes the k=0 case of Galves-Schmitt.

2. Generating Function Method

Using probability generating functions F_s(z) = ∑_^∞ μ^s_k z^k, controlling error terms through Cauchy estimates:

Lemma 5.2: There exists a constant C such that: $\left|\frac{1}{k!}\frac{d^k}{dz^k}E^r_s(0)\right| ≤ Ckrη̃(s)$

3. Chen-Stein Method

Employing the Chen-Stein method to establish compound Poisson approximation, which is more suitable for handling dependent structures compared to traditional moment methods.

Experimental Setup

Theoretical Verification

The paper is primarily theoretical work, verified through:

1. Gibbs-Markov Measures: Verifying the theory's application in Gibbs-Markov systems
2. Gauss Map: Using the Gauss map Ω = (0,1], Tx = 1/x mod 1 as a concrete example
3. Partition Structure: A = {(1/(j+1), 1/j]: j ∈ ℕ}

Parameter Conditions

• φ-mixing decay rate: φ(x) = O(x^{-p}), p > 0
• Observation time scale: N_n = t/(α_1μ(U_n)) (when extreme value index α_1 > 0)
• Block length selection: Δ_n = s^β_n, β ∈ (0,1)

Experimental Results

Main Results

1. φ-Mixing Superior to α-Mixing

Theorem 6.1 demonstrates that the α-mixing case requires the additional assumption that λ_k = lim_{n→∞} λ_k(s_n, U_n) exists, whereas the φ-mixing case automatically guarantees this limit exists by Theorem 2.1.

2. Kac Scale Adjustment

When the extreme value index α_1 > 0, the observation time is: $N_n = \frac{t}{α_1μ(U_n)}$ This provides a speed adjustment to the traditional Kac scale.

3. Compound Poisson Parameters

The clustering probability parameters satisfy: $λ_k = \frac{α_k - α_{k+1}}{α_1}$ where α_k = lim_{L→∞} lim_{n→∞} P(τ^{k-1} < L ≤ τ^k|U_n).

Technical Estimates

Lemma 2.3: Under φ-mixing conditions, for γ ∈ (0, 1/2): $rP(Z^L_U ≥ 1) ≤ 4P(Z^{rL}_U ≥ 1)$

Lemma 2.4: The error estimate is: $\left|\frac{P(Z^{rL}_U = k)}{rP(Z^L_U = k)} - 1\right| ≤ C\left(P(Z^{rL}_U ≥ 1) + φ(L^β) + L^{-(1-β)}\frac{1}{λ_k(rL, U)}\right)$

Related Work

Historical Development

1. Pitskel (1991): Poisson limits for axiom A maps
2. Hirata (1993): Spectral methods and weak mixing properties
3. Galves-Schmitt (1997): Exponential distributions for ψ-mixing
4. Abadi's Series of Works: Extensions to α-mixing and φ-mixing
5. Haydn-Vaienti: Pólya-Aeppli distributions for ψ-mixing systems

Positioning of This Work

• First systematic study of higher-order return times for φ-mixing measures
• Providing weaker mixing conditions than α-mixing
• Unified treatment of periodic and non-periodic point cases

Conclusions and Discussion

Main Conclusions

1. Compound Poisson Limit: Proving compound Poisson distribution limits for higher-order return times in φ-mixing systems
2. Mixing Condition Optimization: φ-mixing allows slower decay rates compared to α-mixing
3. Unified Framework: Providing a general method for handling arbitrary zero-measure sets

Limitations

1. Mixing Conditions: Still requires φ-mixing conditions, which may be too strong for certain systems
2. Decay Rate Requirements: Requires polynomial decay φ(x) = O(x^{-p})
3. Technical Assumptions: Certain lemmas require additional technical conditions

Future Directions

1. Weaker Mixing Conditions: Exploring more general dependence structures
2. Non-stationary Cases: Extending to random dynamical systems
3. Application Domains: Applications in extreme value theory and statistical physics

In-Depth Evaluation

Strengths

1. Theoretical Depth: Providing a complete theoretical framework for return times in φ-mixing systems
2. Technical Innovations:
   • Single limit theorem avoiding technical difficulties of double limits
   • Generalized convolution formula having independent value
   • Clever application of Chen-Stein method
3. Unification: Simultaneously handling periodic and non-periodic cases
4. Practicality: Providing concrete applications to important categories such as Gibbs-Markov systems

Weaknesses

1. Proof Complexity: Proofs of certain technical lemmas are quite involved
2. Condition Restrictions: Verification of φ-mixing conditions in practical systems may be difficult
3. Lack of Numerical Examples: Primarily theoretical work lacking concrete numerical verification

Impact

1. Theoretical Contribution: Providing important supplements to statistical theory of dynamical systems
2. Methodological Value: Developed techniques applicable to other related problems
3. Application Prospects: Potential applications in rare event analysis and extreme value theory

Applicable Scenarios

1. Dynamical Systems Theory: Studying statistical properties of measure-preserving systems
2. Probability Theory: Limit theorems for dependent stochastic processes
3. Statistical Physics: Statistical laws of system return to equilibrium
4. Financial Mathematics: Statistical modeling of extreme events

References

The paper cites important works in the field, including:

• Pitskel's pioneering work
• Galves-Schmitt methodology
• Abadi's contributions on mixing processes
• Haydn-Vaienti's work on distributions near periodic orbits
• Melbourne-Nicol's work on non-uniformly hyperbolic systems