A characterization of generalized Lipschitz classes by the rate of convergence of semi-discrete operators

Basic Information

• Paper ID: 2502.11952
• Title: A characterization of generalized Lipschitz classes by the rate of convergence of semi-discrete operators
• Authors: Danilo Costarelli, Michele Piconi, Gianluca Vinti (University of Perugia, Italy)
• Classification: math.FA (Functional Analysis)
• Publication Date: February 17, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2502.11952

Abstract

This paper establishes a complete characterization of generalized Lipschitz classes by studying the convergence rate of a family of Durrmeyer-type semi-discrete sampling operators in $L^p$ spaces. The research provides direct approximation results with quantitative estimates obtained through K-functionals in Sobolev spaces and higher-order moduli of smoothness. Furthermore, Hardy-Littlewood maximal inequality methods are introduced to weaken the assumptions on kernel functions. Through direct and inverse theorems, a characterization of Lipschitz classes based on the $L^p$ convergence rate of Durrmeyer sampling operators is established. Finally, applications are given for kernel combinations satisfying Strang-Fix type conditions, which simultaneously improve convergence rates and enable signal prediction.

Research Background and Motivation

Problem to be Solved

This paper addresses a fundamental problem in function approximation theory: how to completely characterize the regularity of functions (i.e., generalized Lipschitz classes) through the convergence rate of operators. Specifically, it studies the equivalence relationship between the convergence rate of Durrmeyer-type sampling operators in $L^p$ spaces and the Lipschitz classes to which the approximated functions belong.

Importance of the Problem

1. Theoretical Significance: Establishing precise correspondence between approximation rate and function regularity is a core problem in approximation theory, providing deep insights into the structure of function spaces.
2. Practical Applications: Durrmeyer sampling operators are fundamental mathematical models in signal and image processing; understanding their approximation properties is crucial for practical applications.
3. Historical Continuity: Building upon classical results for Bernstein polynomials, extending them to more general function spaces and operator families.

Limitations of Existing Methods

1. Classical Bernstein Polynomials: While having good approximation properties for continuous functions, they are only applicable to bounded intervals $[0,1]$ and cannot handle discontinuous functions.
2. Existing Durrmeyer Operator Research: Most studies focus on direct approximation theorems; inverse theorems (deriving regularity from convergence rate) are less studied and require stringent conditions.
3. Kernel Function Assumptions: Previous research typically requires strong conditions such as compact support of kernels, limiting the scope of applications.

Research Motivation of This Paper

1. Establish a complete approximation theory in the general framework of $L^p(\mathbb{R})$ spaces ($1 \leq p \leq +\infty$)
2. Improve approximation order through Strang-Fix type vanishing moment conditions
3. Weaken the compact support assumption on kernel functions using Hardy-Littlewood maximal inequalities
4. Establish a complete system of direct and inverse theorems, achieving equivalent characterization of Lipschitz classes

Core Contributions

The main contributions of this paper include:

1. Direct Approximation Theorems: For different values of $p$, convergence rate estimates for Durrmeyer sampling operators are established (Theorems 3.1-3.3), proving that for $f \in L^p(\mathbb{R})$: $\|D^{\phi,\psi}_w f - f\|_p \leq C \cdot \omega_r\left(f, \frac{1}{w}\right)_p$ where $\omega_r$ is the $r$-th order modulus of smoothness.
2. Weakened Kernel Function Assumptions: Through Hardy-Littlewood maximal inequalities (Theorem 3.3), the compact support assumption on the continuous kernel $\psi$ is avoided, requiring only decay conditions $\psi(u) = O(|u|^{-\theta})$.
3. Inverse Approximation Theorems: It is proven that from the convergence rate $\|D^{\phi,\psi}_w f - f\|_p = O(w^{-\alpha})$, one can derive $f \in \text{Lip}^*(\alpha, L^p)$ (Theorem 4.1), which is the technically most difficult part.
4. Complete Characterization Theorem: An equivalent characterization of Lipschitz classes is established (Theorem 4.2): $f \in \text{Lip}^*(\alpha, L^p) \Longleftrightarrow \|D^{\phi,\psi}_w f - f\|_p = O(w^{-\alpha})$
5. Practical Applications: Concrete examples such as B-spline kernels and Jackson kernels are provided, demonstrating how to improve approximation order and achieve signal prediction through linear combinations of kernels.

Detailed Methodology

Task Definition

Research Object: Durrmeyer-type sampling operators defined as $(D^{\phi,\psi}_w f)(x) = \sum_{k \in \mathbb{Z}} \phi(wx - k) w \int_{\mathbb{R}} \psi(wu - k) f(u) du, \quad x \in \mathbb{R}$

where:

• $w > 0$ is the sampling parameter
• $\phi$: discrete kernel function
• $\psi$: continuous kernel function
• $f \in L^p(\mathbb{R})$: function to be approximated

Objective: Establish the equivalence relationship between the convergence rate $\|D^{\phi,\psi}_w f - f\|_p$ and the regularity (Lipschitz class) of $f$.

Core Mathematical Framework

1. Function Space System

Sobolev Spaces: $W^{r,p}(\mathbb{R})$ consists of functions satisfying:

• $f \in L^p(\mathbb{R})$
• Distributional derivative $f^{(r-1)} \in AC(\mathbb{R})$ (absolutely continuous)
• $f^{(r)} \in L^p(\mathbb{R})$

Generalized Lipschitz Spaces: For $\alpha = r + \beta$ ($r \in \mathbb{N}_0$, $0 < \beta \leq 1$):

• $\text{Lip}(\alpha, L^p)$: $f^{(r)} \in \text{Lip}(\beta, L^p)$
• $\text{Lip}^*(\alpha, L^p)$: defined through $r$-th order modulus of smoothness, $\omega_r(f, \delta)_p = O(\delta^\alpha)$

Key Relationships:

• When $\alpha$ is non-integer: $\text{Lip}^*(\alpha, L^p) = \text{Lip}(\alpha, L^p)$
• When $\alpha = r \in \mathbb{N}$: $\text{Lip}(r, L^p) = W^{r,p}(\mathbb{R})$ ($p > 1$)

2. Key Conditions on Kernel Functions

(i) Algebraic Moment Independence: For $\nu = 1, \ldots, r$, the discrete algebraic moments $m_\nu(\phi, u) = \sum_{k \in \mathbb{Z}} \phi(u - k)(k - u)^\nu$ are independent of $u$, denoted as $m_\nu(\phi)$.

(ii) Strang-Fix Type Vanishing Moment Conditions: For $i = 1, \ldots, r-1$, $\sum_{\nu=0}^i \binom{i}{\nu} m_{i-\nu}(\phi) \tilde{m}_\nu(\psi) = 0$

where $\tilde{m}_\nu(\psi) = \int_{\mathbb{R}} u^\nu \psi(u) du$ are continuous algebraic moments.

Physical Meaning: These conditions ensure that lower-order terms in Taylor expansion vanish, thereby improving the approximation order.

3. K-Functionals and Moduli of Smoothness

K-Functional: $K(f, t; L^p, W^{r,p}) = \inf_{g \in W^{r,p}(\mathbb{R})} \left\{ \|f - g\|_p + t\|g^{(r)}\|_p \right\}$

Equivalence Relation (Johnen's Theorem 2.1): There exist constants $C_1, C_2$ such that $C_1 \omega_r(f, t)_p \leq K(f, t^r; L^p, W^{r,p}) \leq C_2 \omega_r(f, t)_p$

This equivalence serves as a bridge connecting approximation rate with function regularity.

Proof Strategy for Direct Theorems

Taylor Expansion and Vanishing Moments

Using Taylor's formula with integral remainder: $f(u) = f(x) + \sum_{i=1}^{r-1} \frac{f^{(i)}(x)}{i!}(u-x)^i + \int_x^u \frac{f^{(r)}(t)}{(r-1)!}(u-t)^{r-1} dt$

Applying to $(D^{\phi,\psi}_w f)(x)$ and using vanishing moment condition (ii), we obtain the key simplification (Equation 4): $(D^{\phi,\psi}_w g)(x) = g(x) + \sum_{k \in \mathbb{Z}} \phi(wx-k) w \int_{\mathbb{R}} \psi(wu-k) \left\{ \int_x^u \frac{g^{(r)}(t)}{(r-1)!}(u-t)^{r-1} dt \right\} du$

Three Technical Approaches

Approach 1 (Theorem 3.1, $p = \infty$):

• Direct estimation of integral remainder
• Requires decay condition $\theta > r+1$

Approach 2 (Theorem 3.2, $1 \leq p < \infty$, compact support $\psi$):

• Three applications of Jensen's inequality
• Refined estimation by dividing integration domain into $I_1$ and $I_2$
• Using Fubini-Tonelli theorem to exchange order of integration

Approach 3 (Theorem 3.3, $1 < p < \infty$, general $\psi$):

• Introduction of Hardy-Littlewood maximal function: $Mf(x) = \sup_{u \neq x} \frac{1}{|x-u|} \left| \int_x^u f(t) dt \right|$
• Using maximal inequality $\|Mf\|_p \leq C_p \|f\|_p$
• Avoiding compact support assumption

Proof Strategy for Inverse Theorem (Theorem 4.1)

This is the technically most difficult part, employing mathematical induction:

Preparatory Lemma (Lemma 4.1)

For $\phi \in W^{r,1}(\mathbb{R})$, the algebraic moments of its distributional derivatives satisfy: $m_\mu(\phi^{(\nu)}, u) = \begin{cases} 0, & \nu \neq \mu \\ \nu!, & \nu = \mu \end{cases}$

Inductive Proof Structure

Assumption: $\|D^{\phi,\psi}_w f - f\|_p = O(w^{-\alpha})$, $0 < \alpha < r$

Goal: Prove $\omega_m(f, \delta)_p = O(\delta^\alpha)$, where $m$ is the integer satisfying $m-1 \leq \alpha < m$

Key Steps:

1. Fix $\delta = 1/w$ and decompose using triangle inequality: $\omega_m(f, \delta)_p \leq \omega_m(f - D^{\phi,\psi}_w f, \delta)_p + \omega_m(D^{\phi,\psi}_w f, \delta)_p$
2. Use the regularity $D^{\phi,\psi}_w f \in W^{r,p}(\mathbb{R})$ to estimate the second term through K-functionals
3. Compute $(D^{\phi,\psi}_w f)^{(m)}$ and use Lemma 4.1 to eliminate lower-order terms
4. Establish recursive inequality (Equation 10): $\omega_m(f, \delta_n)_p \leq \frac{K_1}{w^\alpha} + K_2 \delta_n^m w^m \omega_m(f, w^{-1})_p$
5. Choose $A > \bar{w} > 1$ such that $2K_2 < A^{m-\alpha}$, perform induction on $\delta_n = A^{-n}$
6. Complete the inductive step through "either-or" argument (either $a \leq 2b$ or $a \leq 2c$)

Experimental Setup

This is a pure theoretical mathematics paper without traditional experiments. However, Section 5 provides concrete examples of theoretical applications.

Kernel Function Instances

1. Central B-Spline (Equation 12)

$\sigma_n(u) = \frac{1}{(n-1)!} \sum_{j=0}^n (-1)^j \binom{n}{j} \left( \frac{n}{2} + u - j \right)_+^{n-1}$

Properties:

• Fourier transform: $\hat{\sigma}_n(v) = \text{sinc}^n(v/(2\pi))$
• Compact support: $\text{supp}(\sigma_n) = [-n/2, n/2]$
• Algebraic moments: Through Poisson summation formula, $m_\nu(\sigma_n, x) = \tilde{m}_\nu(\sigma_n)$ for $\nu = 1, \ldots, n-1$

2. Jackson Kernel

$J_{N,\alpha}(u) = c_{N,\alpha} \text{sinc}^{2N}\left( \frac{u}{2N\pi\alpha} \right)$

Properties:

• Non-compact support but fast decay: $J_{N,\alpha}(u) = O(|u|^{-2N})$
• Fourier transform has compact support on $[-1/\alpha, 1/\alpha]$

3. Linear Combination Kernel (Figure 1)

$\tau(u) = \frac{1}{8} \{ 47\sigma_3(u-2) - 62\sigma_3(u-3) + 23\sigma_3(u-4) \}$

Features:

• Coefficient design makes $r=3$ order moments vanish
• Support on $(0, +\infty)$, achieving predictive property: predicting current value using only past samples

Summary of Application Results

Corollary 5.1 (B-splines, $n \geq 2$): $\|D^{\sigma_n, \sigma_n}_w f - f\|_p \leq C \cdot \omega_2(f, 1/w)_p$

Corollary 5.2 (Lipschitz class characterization): $f \in \text{Lip}^*(\alpha, L^p) \Leftrightarrow \|D^{\sigma_n, \sigma_n}_w f - f\|_p = O(w^{-\alpha}), \quad 0 < \alpha < 2$

Experimental Results

Main Theoretical Results

Theorem 3.1 ($L^\infty$ Case)

Under condition $\theta > r+1$, for any $f \in C(\mathbb{R})$: $\|D^{\phi,\psi}_w f - f\|_\infty \leq \Lambda_\infty \cdot \omega_r(f, 1/w)_\infty$

Theorem 3.2 ($L^p$ Case, Compact Support)

For $1 \leq p < \infty$, $\psi$ with compact support, $\theta > pr+1$: $\|D^{\phi,\psi}_w f - f\|_p \leq \Lambda_p \cdot \omega_r(f, 1/w)_p$

Theorem 3.3 ($L^p$ Case, General Kernel)

For $1 < p < \infty$, $\theta > pr+1$ (without compact support assumption): $\|D^{\phi,\psi}_w f - f\|_p \leq \Gamma_p \cdot \omega_r(f, 1/w)_p$

Theorem 4.2 (Equivalent Characterization)

Under appropriate conditions, for $0 < \alpha < r$: $f \in \text{Lip}^*(\alpha, L^p) \Longleftrightarrow \|D^{\phi,\psi}_w f - f\|_p = O(w^{-\alpha})$

Result Analysis

1. Approximation Order: Through Strang-Fix conditions, high-order approximation $O(w^{-r})$ can be achieved
2. Space Dependence:
   • $p=1$: requires $\psi$ with compact support
   • $1 < p < \infty$: Hardy-Littlewood method can avoid compact support
   • $p = \infty$: weakest conditions
3. Decay Requirements:
   • Discrete kernel: $\theta > r+1$ (basic)
   • Continuous kernel: $\theta > pr+1$ ($L^p$ case more stringent)
4. Lipschitz Class Equivalence:
   • Non-integer $\alpha$: $\text{Lip}^*(\alpha, L^p) = \text{Lip}(\alpha, L^p)$
   • Integer $r$: $\text{Lip}(r, L^p) = W^{r,p}(\mathbb{R})$ ($p>1$)

Related Work

Historical Context

1. Bernstein Polynomials (1912):
   • Classical result: $|(B_n f)(x) - f(x)| \leq M(x(1-x)/n)^{\alpha/2}$
   • Inverse theorems: Berens, Lorentz, MacKenzie (1972), Becker (1978)
2. Durrmeyer Method (1967):
   • Replacing point values with integrals: $(D_n f)(x) = (n+1) \sum_{\nu=0}^n p_{\nu,n}(x) \int_0^1 p_{\nu,n}(u) f(u) du$
   • Extension to $L^p$ spaces
3. Generalized Sampling Theory:
   • Butzer school (1980s): establishing sampling operator theory
   • Strang-Fix conditions: key to improving approximation order

Relationship of This Paper to Related Work

Compared to Classical Bernstein Theory:

• Extension to entire real axis $\mathbb{R}$
• Handling $L^p$ functions rather than only continuous functions
• Establishing higher-order approximation theory

Compared to Existing Durrmeyer Operator Research:

• First complete inverse theorem (from convergence rate to regularity)
• Introduction of Hardy-Littlewood method to weaken assumptions
• Systematic study of Lipschitz class characterization

Compared to Kantorovich Operators:

• Kantorovich is a special case ($\psi = \chi_{[0,1]}$)
• This paper provides more general framework

Conclusions and Discussion

Main Conclusions

1. Complete Characterization: Establishes necessary and sufficient conditions between convergence rate of Durrmeyer sampling operators and generalized Lipschitz classes
2. Methodological Contributions:
   • Direct theorems: Taylor expansion + vanishing moments + K-functionals
   • Inverse theorems: induction + distributional derivative moment properties
   • Hardy-Littlewood method: weakening kernel function assumptions
3. Application Value:
   • Provides kernel design methods for high-order approximation
   • Enables signal prediction functionality
   • Applicable to image and signal processing

Limitations

1. $p=1$ Case: Hardy-Littlewood maximal inequality fails in $L^1$, still requiring compact support assumption
2. Integer Order Case: When $\alpha = r \in \mathbb{N}$, $\text{Lip}(r, L^p) \subsetneq \text{Lip}^*(r, L^p)$, characterization is incomplete
3. Kernel Function Design: Finding kernels satisfying Strang-Fix conditions requires detailed computation
4. Computational Complexity: Paper does not discuss numerical implementation and computational efficiency of operators

Future Directions

1. Multivariate Case: Extension to $\mathbb{R}^d$ (authors have preliminary work 23)
2. Weighted Spaces: Study properties in weighted $L^p$ spaces
3. Nonlinear Generalizations: Consider nonlinear sampling operators
4. Numerical Algorithms: Develop efficient computational methods
5. Practical Applications: Concrete implementations in image processing and signal reconstruction

In-Depth Evaluation

Strengths

1. Theoretical Completeness:
   • Direct theorems + inverse theorems form complete system
   • Covers all cases $1 \leq p \leq \infty$
   • Technically rigorous with detailed proofs
2. Methodological Innovation:
   • Introduction of Hardy-Littlewood method is a highlight, significantly weakening assumptions
   • Inductive proof technique for inverse theorem is clever, particularly utilizing distributional derivative moment properties
3. Practical Value:
   • Provides concrete usable kernel function examples (B-splines, Jackson kernels, linear combinations)
   • Predictive property has direct applications in signal processing
4. Writing Quality:
   • Clear structure, natural transition from classical results to new results
   • Precise mathematical exposition, well-organized lemmas and theorems

Weaknesses

1. Technical Assumptions:
   • Decay condition $\theta > pr+1$ requires stronger conditions when $p$ is large
   • Verification of vanishing moment conditions requires extensive computation
2. Absence of Experiments:
   • As a pure theory paper, lacks numerical experiments validating theoretical results
   • No practical signal processing case studies provided
3. $L^1$ Case:
   • Theorem 3.2 still requires compact support assumption, inconsistent with other $p$ values
   • This is an inherent limitation of Hardy-Littlewood method
4. Computational Aspects:
   • Does not discuss explicit estimates for constants $\Lambda_p$, $\Gamma_p$
   • Does not analyze numerical stability of operators

Impact and Influence

1. Theoretical Contribution:
   • Resolves long-standing challenge in inverse approximation problem for Durrmeyer sampling operators
   • Provides new tools and perspectives for function approximation theory
2. Application Prospects:
   • Provides theoretical foundation for signal processing
   • Predictive property has potential applications in real-time systems
3. Reproducibility:
   • Theoretical proofs are complete and verifiable
   • Kernel function examples are concrete and explicit
   • However, lacks code implementation
4. Follow-up Research:
   • May inspire deeper research in multivariate cases
   • Motivates similar research for other operator families

Applicable Scenarios

1. Mathematical Research:
   • Function approximation theory
   • Harmonic analysis
   • Sobolev space theory
2. Signal Processing:
   • Sampling theory
   • Signal reconstruction
   • Real-time prediction systems
3. Image Processing:
   • Image approximation
   • Edge detection (using regularity characterization)
4. Numerical Analysis:
   • Numerical integration
   • Function interpolation

Key References

17 P. L. Butzer and R. J. Nessel, Fourier analysis and approximation, Vol. 1, 1971. (Classical textbook)

29 R. A. DeVore and G. G. Lorentz, Constructive approximation, 1993. (Sobolev spaces and K-functional theory)

34 H. Johnen, Inequalities connected with the moduli of smoothness, 1972. (K-functional and modulus of smoothness equivalence)

39 E. M. Stein, Singular integrals and differentiability properties of functions, 1970. (Hardy-Littlewood maximal inequality)

18-20 P. L. Butzer et al., Series of papers on generalized sampling theory (establishing sampling operator theory framework)

Overall Assessment: This is a high-quality theoretical paper in function approximation theory, achieving significant progress in studying approximation properties of Durrmeyer sampling operators. The proof of the inverse theorem and the introduction of Hardy-Littlewood method are particularly innovative. The paper is technically rigorous and theoretically complete, providing valuable tools for the field. Main shortcomings include absence of numerical experiments and practical application cases, and incomplete treatment of the $L^1$ case.