2025-11-12T01:58:29.022302

On the Extreme Value Behavior of $\vartheta$-Expansions

Sebe, Lascu, Selmi
The main objective of this paper is to develop extreme value theory for $\vartheta$-expansions. We establish the limit distribution of the maximum value in a $\vartheta$-continued fraction mixing stationary stochastic process, along with some related results. These findings are analogous to the theorems of J. Galambos and W. Philipp for regular continued fractions. Additionally, we emphasize that a Borel-Bernstein type theorem plays a crucial role.
academic

On the Extreme Value Behavior of ϑ\vartheta-Expansions

Basic Information

  • Paper ID: 2309.12654
  • Title: On the Extreme Value Behavior of ϑ\vartheta-Expansions
  • Authors: Gabriela Ileana Sebe, Dan Lascu, Bilel Selmi
  • Classification: math.PR (Probability Theory), math.NT (Number Theory)
  • Publication Date: September 2023 (arXiv v2 updated November 1, 2025)
  • Paper Link: https://arxiv.org/abs/2309.12654

Abstract

The primary objective of this paper is to establish extreme value theory for ϑ\vartheta-expansions. The authors establish the limiting distributions of maxima in ϑ\vartheta-continued fraction mixed stationary random processes and related results. These findings parallel theorems by J. Galambos and W. Philipp concerning regular continued fractions. Furthermore, the authors emphasize that Borel-Bernstein type theorems play a crucial role in the investigation.

Research Background and Motivation

Research Questions

This paper investigates the extreme value behavior of partial quotients in ϑ\vartheta-expansions, particularly the asymptotic distribution of the maximum partial quotient LN=max⁡1≤n≤Nbn(x)L_N = \max_{1 \leq n \leq N} b_n(x).

Significance of the Problem

  1. Theoretical Significance: ϑ\vartheta-expansions are an important generalization of regular continued fractions (RCF), introduced by Bhattacharya and Goswami in the context of random number generation. For the special case ϑ=1\vartheta=1, ϑ\vartheta-expansions reduce to classical continued fraction expansions.
  2. Historical Background: For regular continued fractions, foundational results on the extreme value behavior of partial quotients were established by Galambos (1970s) and Philipp (1976). However, such theory has not yet been developed for ϑ\vartheta-expansions.
  3. Applied Value: The growth rate of partial quotients in continued fractions directly affects the precision of rational approximations. According to inequality (1.9): 1qn(x)(qn+1(x)+ϑqn(x))≤∣x−pn(x)qn(x)∣≤1qn(x)qn+1(x)\frac{1}{q_n(x)(q_{n+1}(x) + \vartheta q_n(x))} \leq \left|x - \frac{p_n(x)}{q_n(x)}\right| \leq \frac{1}{q_n(x)q_{n+1}(x)} The approximation precision depends on the growth of the denominator qnq_n, which is determined recursively by the partial quotients bnb_n.

Limitations of Existing Methods

  • Classical Galambos-Philipp theory applies only to regular continued fractions (ϑ=1\vartheta=1)
  • ϑ\vartheta-expansions possess different invariant measures and mixing properties
  • New techniques are needed to handle the complexity introduced by the parameter ϑ\vartheta

Research Motivation

To establish a complete extreme value theory framework for ϑ\vartheta-expansions, generalize classical results to more general settings, and reveal the influence of the parameter ϑ\vartheta on extreme value behavior.

Core Contributions

The main contributions of this paper include:

  1. Borel-Bernstein Type Theorem (Theorem 3.1): Establishes a "0-1" law for the growth rate of partial quotients in ϑ\vartheta-expansions, which serves as the foundation for subsequent extreme value theory.
  2. Fréchet Limit Law (Theorem 4.5): Proves that the standardized limiting distribution of the maximum partial quotient follows a Fréchet distribution: lim⁡N→∞γϑ(LN(x)<Nylog⁡(1+ϑ2))=exp⁡(−1y)\lim_{N \to \infty} \gamma_\vartheta\left(L_N(x) < \frac{Ny}{\log(1+\vartheta^2)}\right) = \exp\left(-\frac{1}{y}\right)
  3. Convergence Rate Estimates (Theorem 4.6): Provides the convergence rate of the limiting distribution: γϑ(LN(x)<Nylog⁡(1+ϑ2))=exp⁡(−1y)+O(exp⁡(−(log⁡N)δ))\gamma_\vartheta\left(L_N(x) < \frac{Ny}{\log(1+\vartheta^2)}\right) = \exp\left(-\frac{1}{y}\right) + O\left(\exp\left(-(\log N)^\delta\right)\right)
  4. Iterated Logarithm Law (Theorem 5.2): Establishes the precise growth rate of the maximum partial quotient: lim inf⁡N→∞LN(x)log⁡log⁡NN=1log⁡(1+ϑ2)a.e.\liminf_{N \to \infty} \frac{L_N(x) \log \log N}{N} = \frac{1}{\log(1+\vartheta^2)} \quad \text{a.e.}
  5. Mixing Properties (Lemma 4.4): Proves that the partial quotient sequence in ϑ\vartheta-expansions satisfies ψ\psi-mixing conditions with exponential decay rate.

Methodology Details

Task Definition

Input: Fixed parameter ϑ∈(0,1)\vartheta \in (0,1) satisfying ϑ2=1/m\vartheta^2 = 1/m (m∈N+m \in \mathbb{N}^+), and an irrational number xx on the interval [0,ϑ][0,\vartheta].

Output: The probability distribution and asymptotic behavior of LN(x)=max⁡1≤n≤Nbn(x)L_N(x) = \max_{1 \leq n \leq N} b_n(x), the maximum of the first NN partial quotients in the ϑ\vartheta-expansion of xx.

Constraints:

  • bn(x)≥mb_n(x) \geq m for all n∈N+n \in \mathbb{N}^+
  • Consider the invariant measure γϑ\gamma_\vartheta: dγϑ=1log⁡(1+ϑ2)⋅ϑdx1+ϑxd\gamma_\vartheta = \frac{1}{\log(1+\vartheta^2)} \cdot \frac{\vartheta dx}{1+\vartheta x}

Theoretical Framework

1. Basic Properties of ϑ\vartheta-Expansions

For x∈(0,ϑ)x \in (0,\vartheta), define the generalized Gauss map:

\frac{1}{x} - \vartheta\left\lfloor\frac{1}{x\vartheta}\right\rfloor, & x \in (0,\vartheta] \\ 0, & x = 0 \end{cases}$$ The $\vartheta$-expansion of $x$ is: $$x = [b_1\vartheta, b_2\vartheta, b_3\vartheta, \ldots] = \cfrac{1}{b_1\vartheta + \cfrac{1}{b_2\vartheta + \cfrac{1}{b_3\vartheta + \cdots}}}$$ where $b_n(x) = b_1(T_\vartheta^{n-1}(x))$. #### 2. Measure Estimates for Cylinder Sets For $n$-th order cylinder sets $C(b_1, \ldots, b_n) = \{x \in \Omega: b_k(x) = i_k, k=1,\ldots,n\}$, there is a key estimate (formula 2.4): $$\frac{1}{2q_n^2} \leq \frac{1}{(1+\vartheta^2)q_n^2} \leq \lambda_\vartheta(C(b_1,\ldots,b_n)) \leq \frac{1}{q_n^2}$$ and relative measure estimates (Lemma 2.1): $$\frac{1}{6k^2} < \lambda_\vartheta\left(\bigcup_{b_1,\ldots,b_n \geq m} C(b_1,\ldots,b_n,k)\right) < \frac{m+1}{k^2}$$ ### Core Technical Methods #### Method 1: Proof Strategy for Borel-Bernstein Type Theorem **Theorem 3.1**: For a function $\phi: \mathbb{N}^+ \to (0,+\infty)$, define $$A_\phi = \{x \in \Omega: b_n(x) > \phi(n) \text{ for infinitely many } n \in \mathbb{N}^+\}$$ Then: $$\lambda_\vartheta(A_\phi) = \begin{cases} 0, & \text{if } \sum_{n \geq 1} \frac{1}{\phi(n)} < \infty \\ 1, & \text{if } \sum_{n \geq 1} \frac{1}{\phi(n)} = \infty \end{cases}$$ **Proof Strategy**: 1. **Convergence Case**: Apply the Borel-Cantelli lemma. Set $A_n = \{x: b_n(x) > \phi(n)\}$. By Lemma 2.1: $$\lambda_\vartheta(A_n) < \frac{2(m+1)}{\phi(n)}$$ If $\sum \frac{1}{\phi(n)} < \infty$, then $\sum \lambda_\vartheta(A_n) < \infty$, hence $\lambda_\vartheta(A_\phi) = 0$. 2. **Divergence Case**: We need to prove $\lambda_\vartheta(A_\phi^c) = 0$. Define $B_{j,\ell} = \bigcap_{j < n \leq j+\ell} A_n^c$. By induction: $$\lambda_\vartheta(B_{j,\ell}) \leq \prod_{i=1}^\ell \left(1 - \frac{1}{12\phi(j+i)}\right) \leq \exp\left(-\sum_{i=1}^\ell \frac{1}{12\phi(j+i)}\right)$$ When $\sum \frac{1}{\phi(n)} = \infty$, we have $\lim_{\ell \to \infty} \lambda_\vartheta(B_{j,\ell}) = 0$. #### Method 2: Mixing Properties and Extreme Value Distribution **$\psi$-Mixing Condition** (Lemma 4.4): For $A \in \mathcal{B}_1^k$ and $B \in \mathcal{B}_{k+n}^\infty$: $$|\gamma_\vartheta(A \cap B) - \gamma_\vartheta(A)\gamma_\vartheta(B)| \leq K_\vartheta q_\vartheta^n \gamma_\vartheta(A)\gamma_\vartheta(B)$$ where $0 < q_\vartheta < 1$ is the exponential decay rate. **Proof Technique**: - Construct recursive function sequences $\{f_{\vartheta,n}\}$ - Utilize smoothness estimates of conditional distribution functions - Establish uniform bounds on derivatives via Lemma 4.3 **Poincaré Expansion of Extreme Value Distribution** (Theorem 4.5): Set $B_N = \{x: L_N(x) < w\}$. Using the inclusion-exclusion principle: $$\gamma_\vartheta(B_N) = \sum_{k=0}^N (-1)^k S_k$$ where $S_k = \sum_{1 \leq n_1 < \cdots < n_k \leq N} \gamma_\vartheta(A_{n_1} \cap \cdots \cap A_{n_k})$. Key steps: 1. **Tail Estimates**: Use mixing properties to control the contribution of large $k$ terms 2. **Principal Term Analysis**: For fixed $k < Z$, separate the principal term $S_k^*$ from the remainder $R_k$ 3. **Asymptotic Expansion**: Prove that $S_k = \frac{y^{-k}}{k!} + o(1)$ ### Technical Innovations 1. **Handling Parameterized Invariant Measures**: Unlike the classical Gauss measure, $\gamma_\vartheta$ depends on the parameter $\vartheta$, requiring new measure estimates. 2. **Precise Control of Mixing Speed**: By constructing special function classes (Lemma 4.3), the authors prove exponential mixing speed with constant $q_\vartheta$ satisfying: $$q_\vartheta = m\sum_{i \geq m} \left(\frac{m}{i^3(i+1)} + \frac{i+1-m}{i(i+1)^3}\right) < 1$$ 3. **Optimization of Convergence Rate**: Theorem 4.6 achieves logarithmic error bounds by carefully choosing parameters $Z = \lfloor\frac{\log N}{\log \log N}\rfloor$ and $t = \lfloor(\log N)^2\rfloor$. 4. **Application of Quantitative Borel-Cantelli Lemma**: In the proof of the iterated logarithm law, Philipp's quantitative version (Lemma 5.1) is used to handle dependent event sequences. ## Experimental Setup **Note**: This is a pure theoretical mathematics paper with no numerical experiments. All results are rigorous mathematical proofs. ### Theoretical Verification Framework The paper verifies the correctness of the theory through: 1. **Special Case Verification**: When $\vartheta \to 1$, the results should reduce to the classical Galambos-Philipp theorem 2. **Internal Consistency**: - Corollary 3.2 of Theorem 3.1 provides specific growth rates - Results of Theorems 4.5 and 4.6 are compatible - Theorems 5.2 and 5.3 mutually support each other 3. **Measure Equivalence**: Robustness of results is verified by utilizing the equivalence of $\lambda_\vartheta$ and $\gamma_\vartheta$ (in the proof of Theorem 5.2) ## Experimental Results ### Main Theoretical Results #### Result 1: Growth Rate of Partial Quotients (Corollary 3.2) For $\lambda_\vartheta$-almost every $x \in [0,\vartheta]$: - **Lower Bound**: $b_n(x) > n \log n$ holds for infinitely many $n$ - **Upper Bound**: For every $\varepsilon > 0$, when $n$ is sufficiently large, $b_n(x) < n(\log n)^{1+\varepsilon}$ **Significance**: Precisely characterizes the typical growth speed of partial quotients as $\Theta(n \log n)$. #### Result 2: Fréchet Limit Law (Theorem 4.5) For any $y > 0$: $$\lim_{N \to \infty} \gamma_\vartheta\left(L_N(x) < \frac{Ny}{\log(1+\vartheta^2)}\right) = \exp\left(-\frac{1}{y}\right)$$ **Interpretation**: - The standardization factor is $\frac{N}{\log(1+\vartheta^2)}$ - The limiting distribution is the Fréchet distribution $\Phi_\alpha(y) = \exp(-y^{-\alpha})$ with $\alpha = 1$ - This indicates that the maximum partial quotient grows at rate $\frac{N}{\log(1+\vartheta^2)}$ #### Result 3: Convergence Rate (Theorem 4.6) For $0 < \delta < 1$ and $y \geq (\log N)^{-\delta}$: $$\gamma_\vartheta\left(L_N < \frac{Ny}{\log(1+\vartheta^2)}\right) = \exp\left(-\frac{1}{y}\right) + O\left(\exp\left(-(\log N)^\delta\right)\right)$$ **Significance**: The convergence rate is doubly exponential, much faster than polynomial rates. #### Result 4: Liminf Law (Theorem 5.2) For almost every $x \in [0,\vartheta]$: $$\liminf_{N \to \infty} \frac{L_N(x) \log \log N}{N} = \frac{1}{\log(1+\vartheta^2)}$$ **Interpretation**: - This is a precise growth rate, not merely an order of magnitude - The constant $\frac{1}{\log(1+\vartheta^2)}$ explicitly depends on the parameter $\vartheta$ #### Result 5: Dichotomy of Limsup (Corollary 5.4) For non-decreasing functions $\{\phi(n)\}$: $$\limsup_{N \to \infty} \frac{L_N(x)}{\phi(N)}$$ is almost surely either 0 or $\infty$, depending on the convergence or divergence of the series $\sum \frac{1}{\phi(n)}$. ### Theoretical Findings 1. **Parameter Dependence**: All characteristic constants in extreme value behavior are $\frac{1}{\log(1+\vartheta^2)}$, explicitly depending on $\vartheta$. 2. **Critical Role of Mixing**: The exponential mixing speed ($q_\vartheta^n$) is central to obtaining precise convergence rates. 3. **Foundational Nature of Borel-Bernstein Theorem**: It is not only independently interesting but also a key tool for proving the iterated logarithm law. 4. **Correspondence with Classical Results**: When $\vartheta = 1$, $\log(1+\vartheta^2) = \log 2$, recovering the classical Galambos-Philipp results. ## Related Work ### Extreme Value Theory for Continued Fractions 1. **Classical Results**: - **Borel (1909) & Bernstein (1912)**: First established "0-1" laws for regular continued fractions - **Galambos (1972, 1973, 1974)**: Proved Fréchet distribution and iterated logarithm law for maximum partial quotients of RCF - **Philipp (1976)**: Resolved Erdős conjecture, providing precise convergence rates 2. **Research on $\vartheta$-Expansions**: - **Bhattacharya & Goswami (2000)**: Introduced the concept of $\vartheta$-expansions - **Chakraborty & Rao (2003)**: Established basic theory, proved ergodicity - **Sebe & Lascu (2014, 2017, 2019)**: Studied Gauss-Kuzmin problems and convergence rates - **This Paper**: First to establish complete extreme value theory ### Extreme Value Theory for Mixing Processes - **Philipp (1967)**: Developed metric theorems in number theory, introduced quantitative Borel-Cantelli lemma (Lemma 5.1) - This paper adapts these techniques to the special structure of $\vartheta$-expansions ### Relative Advantages of This Paper 1. **Completeness**: First systematic establishment of extreme value theory framework for $\vartheta$-expansions 2. **Precision**: Provides not only limiting distributions but also convergence rates 3. **Technical Depth**: Handles new difficulties arising from parameterized invariant measures and mixing properties 4. **Unification**: Includes classical results as special cases ## Conclusions and Discussion ### Main Conclusions 1. **Extreme Value Distribution**: The maximum partial quotient in $\vartheta$-expansions follows a Fréchet distribution with standardization factor $\frac{N}{\log(1+\vartheta^2)}$. 2. **Precise Growth Rate**: Through the iterated logarithm law, the precise growth speed of the maximum partial quotient is established as $\frac{N}{\log \log N \cdot \log(1+\vartheta^2)}$. 3. **Theoretical Completeness**: A theory system completely parallel to that of regular continued fractions is established, demonstrating that Galambos-Philipp theory can be generalized to $\vartheta$-expansions. 4. **Methodological Contribution**: Demonstrates how to handle extreme value problems under parameterized invariant measures; techniques are generalizable to other generalized continued fraction systems. ### Limitations 1. **Parameter Restrictions**: The theory requires $\vartheta^2 = 1/m$ ($m \in \mathbb{N}^+$), which is a sufficient condition for the existence of invariant measures. For general $\vartheta$ values, the existence and properties of invariant measures remain unclear. 2. **Dependence on Mixing Speed**: The convergence rate estimate in Theorem 4.6 depends on the specific value of the mixing coefficient $q_\vartheta$, whose exact computation may be difficult. 3. **Absence of Limsup Law**: As shown in Corollary 5.4, there is no finite nonzero limsup law, contrasting with the liminf law. 4. **Technical Complexity**: The proofs rely on refined analytical techniques (such as recursive function estimates in Lemma 4.3), and generalization to more general settings may require new methods. ### Future Directions 1. **More General Parameters**: Study $\vartheta$ values not satisfying $\vartheta^2 = 1/m$, potentially requiring development of new invariant measure theory. 2. **Multidimensional Generalizations**: Consider extreme value behavior of multidimensional $\vartheta$-expansions. 3. **Dependence Structure**: Deeper investigation of how dependencies among partial quotients affect extreme value behavior. 4. **Applied Research**: - Applications in Diophantine approximation - Quality assessment of random number generation - Chaotic properties of dynamical systems 5. **Numerical Verification**: Although this is pure theoretical work, numerical experiments could provide additional insights. ## In-Depth Evaluation ### Strengths 1. **Theoretical Rigor**: - All theorems have complete mathematical proofs - Clear logical chain, progressing layer by layer from basic lemmas to main theorems - Precise handling of technical details (e.g., derivative estimates in Lemma 4.3) 2. **Originality**: - First to establish complete extreme value theory for $\vartheta$-expansions - Develops new techniques for handling parameterized invariant measures - Proof of mixing properties (Lemma 4.4) has independent value 3. **Completeness**: - Encompasses complete theoretical chain from "0-1" laws to limiting distributions, convergence rates, and iterated logarithm laws - Results mutually support each other, forming a unified theoretical framework 4. **Writing Quality**: - Clear structure, progressing systematically from basic definitions to main results - Sufficient motivation, with clear connections to classical results - Technical details are sufficient but not redundant 5. **Mathematical Depth**: - Synthesizes techniques from multiple fields: probability theory, number theory, ergodic theory, and extreme value theory - Application of Poincaré expansion (Theorem 4.5) demonstrates sophisticated techniques ### Weaknesses 1. **Limited Scope of Applicability**: - Parameter condition $\vartheta^2 = 1/m$ is quite restrictive - Cases with general $\vartheta$ values are not addressed 2. **Absence of Numerical Verification**: - Although a theoretical paper, simple numerical examples could enhance intuitive understanding - No demonstration of specific cases (e.g., $\vartheta = 1/\sqrt{2}$) 3. **Computability of Constants**: - Explicit expressions for mixing constants $K_\vartheta$ and $q_\vartheta$ are not provided - Implicit constants in Theorem 4.6 are not made explicit 4. **Limited Connection to Applications**: - Discussion of practical application scenarios for theoretical results is insufficient - Specific connections to applied fields such as Diophantine approximation could be clearer 5. **Discussion of Technical Obstacles**: - Lacks intuitive explanation for why certain techniques (e.g., specific construction in Lemma 4.3) are necessary ### Impact Assessment 1. **Contribution to the Field**: - **High**: Fills the gap in extreme value theory for $\vartheta$-expansions - Provides foundational framework and technical tools for subsequent research - Demonstrates the generalizability of classical theory 2. **Practical Value**: - **Moderate**: Primarily theoretical contribution with limited direct applications - Potential applications include: * Quality assessment of rational approximations * Theoretical analysis of random number generators * Statistical properties of dynamical systems 3. **Reproducibility**: - **High**: All proofs are complete and verifiable - Technical approach is clear and can be followed by other researchers - All cited lemmas and theorems have clear references 4. **Citation Potential**: - Expected to become a standard reference for $\vartheta$-expansion research - Methodology (particularly handling of mixing properties) may be borrowed by researchers studying other generalized continued fractions ### Applicable Scenarios 1. **Theoretical Research**: - Study of other types of generalized continued fractions (e.g., Lüroth expansions, β-expansions) - Extreme value theory for dynamical systems - Metric theory in number theory 2. **Diophantine Approximation**: - Understanding worst-case approximation quality - Studying approximation properties of specific number types 3. **Algorithm Analysis**: - Complexity analysis of algorithms based on continued fractions - Quality assessment of random number generators 4. **Teaching and Learning**: - Case study of extreme value theory applications in number theory - Demonstration of analytical techniques for mixing processes ## Key References (from the paper) 1. **Borel (1909)** & **Bernstein (1912)**: Foundational work on "0-1" laws for continued fractions 2. **Galambos (1972-1974)**: Series of papers on extreme value theory for RCF 3. **Philipp (1967, 1976)**: Metric theorems and resolution of Erdős conjecture 4. **Chakraborty & Rao (2003)**: Basic theory of $\vartheta$-expansions 5. **Sebe & Lascu (2014-2019)**: Prior research on $\vartheta$-expansions --- **Overall Assessment**: This is a high-quality theoretical mathematics paper that successfully generalizes the classical Galambos-Philipp theory to $\vartheta$-expansions. The technical depth is impressive, particularly in handling parameterized invariant measures and mixing properties, demonstrating innovation. While the scope of applicability is limited by parameter conditions, the theory is complete and profound within its applicable range. The paper establishes a solid foundation for extreme value theory of $\vartheta$-expansions and is expected to become an important reference in this field.