[Paper Review] Divergence of general localized operators on the sets of measure zero

G. A. Karagulyan|ArXiv.org|Dec 8, 2009

Advanced Harmonic Analysis Research5 references3 citations

TL;DR

This paper proves that for any sequence of linear operators with the localization property, and for any set $ E $ of measure zero, there exists a characteristic function $ \mathbb{I}_G $ such that the operator sequence diverges at every point of $ E $, with liminf ≤ 0 and limsup ≥ 1. The result generalizes classical divergence theorems for Fourier series to broad classes of localized operators, including partial sums and linear means with respect to orthogonal systems.

ABSTRACT

We consider sequences of linear operators $U_nf(x)$ with localization property. It is proved that for any set $E$ of measure zero there exists a set $G$ for which $U_n\ZI_G(x)$ diverges at each point $x\in E$. This result is a generalization of analogous theorems known for the Fourier sums operators with respect to different orthogonal systems.

Motivation & Objective

• To establish that divergence of operator sequences on arbitrary sets of measure zero is a universal phenomenon for operators with localization property.
• To extend known results on divergence of Fourier series (e.g., Stechkin, Kahane-Katznelson) to general linear operators with localization.
• To demonstrate that such divergence is not restricted to specific systems but holds for all operators satisfying the localization condition.
• To construct a characteristic function $ \mathbb{I}_G $ such that the operator sequence diverges pointwise on any given null set $ E $.

Proposed method

• Define a sequence of open sets $ G_k $ with regular partitions $ \mathcal{I}_k $, nested and shrinking toward $ E $, ensuring $ E \subset G_k \subset G_{k-1} $.
• Construct a sequence of indices $ \nu(I) $ for intervals $ I \in \mathcal{I}_k $ such that $ U_{\nu(I)}\mathbb{I}_{G_l}(x) \to 1 $ uniformly on $ I $ for $ l \leq k $, and $ U_{\nu(I)}\mathbb{I}_{G_k}(x) \to 0 $ for $ l < k $.
• Use the kernel bound $ |K_n(x,t)| \leq \phi(|x-t|) $ derived from the localization property, where $ \phi $ is a decreasing function.
• Apply the kernel decay and measure control via $ \delta(I) $ to bound contributions from distant intervals and ensure smallness of operator norms.
• Define the set $ G = \bigcup_{i=1}^\infty (G_{2i-1} \setminus G_{2i}) $, creating an alternating sum structure in the operator action.
• Analyze the behavior of $ U_n\mathbb{I}_G(x) $ along subsequences $ \nu(I_{2t}) $ and $ \nu(I_{2t+1}) $, showing liminf = 0 and limsup = 1 at each $ x \in E $.

Experimental results

Research questions

• RQ1 Can the divergence of Fourier-type operators on a given set of measure zero be guaranteed for general localized linear operators?
• RQ2 Is the divergence phenomenon on null sets intrinsic to the localization property, or restricted to specific orthogonal systems?
• RQ3 Can a single characteristic function $ \mathbb{I}_G $ be constructed such that its operator sequence diverges at every point of a given null set $ E $?
• RQ4 What conditions on the kernel $ K_n(x,t) $ ensure that such divergence is unavoidable for any null set $ E $?
• RQ5 Does the localization property alone suffice to generate divergence on any null set, regardless of the function class?

Key findings

• For any set $ E \subset [a,b] $ of measure zero, there exists a measurable set $ G \subset [a,b] $ such that $ \liminf_{n\to\infty} U_n\mathbb{I}_G(x) \leq 0 $ and $ \limsup_{n\to\infty} U_n\mathbb{I}_G(x) \geq 1 $ for all $ x \in E $.
• The construction ensures that along a subsequence $ \nu(I_{2t}) $, $ U_n\mathbb{I}_G(x) \to 0 $, and along $ \nu(I_{2t+1}) $, $ U_n\mathbb{I}_G(x) \to 1 $, proving oscillatory divergence.
• The result holds for all linear operators $ U_n $ with the localization property, including Fourier partial sums and linear means with respect to classical orthogonal systems.
• The function $ f = \mathbb{I}_G $ is not continuous in general, and the divergence cannot be avoided even if the function is bounded.
• The kernel bound $ |K_n(x,t)| \leq \phi(|x-t|) $ with decreasing $ \phi $ is a necessary consequence of the localization property.
• The construction relies on iterative shrinking of open sets $ G_k $ and careful control of measure and kernel decay to ensure operator norms remain bounded and alternating behavior emerges.