Periodic Non-uniqueness Sets for Shift-invariant Spaces and Parity-Based Obstructions to the Frame Property for Gabor Systems

Alexander Ulanovskii, Ilya Zlotnikov · 2026

Read Periodic Non-uniqueness Sets for Shift-invariant Spaces and Parity-Based Obstructions to the Frame Property for Gabor Systems by Alexander Ulanovskii, Ilya Zlotnikov (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-31450.txt.

Introduction

Ulanovskii and Zlotnikov give explicit sets that break the frame property for Gabor systems. A Gabor system shifts and modulates one window function over a time-frequency set, and it is a frame when those copies capture every signal stably. Knowing which sets fail, and why, sharpens the theory's edge cases.

Their periodic non-uniqueness sets, not necessarily lattices, make Gabor systems fail based only on whether the window is even or odd. For many finite-dimensional spaces they also find a window whose Gabor system fails on a high-density lattice. The Gröchenig-Lyubarskii theorem is shown sharp for Gaussian-weighted polynomials: at the critical density an algorithm finds an even or odd polynomial whose system is not a frame. The proofs stay constructive and elementary.

Catalog record

CREATORAlexander Ulanovskii, Ilya Zlotnikov
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.31450

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Periodic Non-uniqueness Sets for Shift-invariant Spaces and Parity-Based Obstructions to the Frame Property for Gabor Systems Alexander Ulanovskii, Ilya Zlotnikov Abstract . The goal of this note is twofold. First, we provide explicit examples of periodic (though not necessarily lattice) sets that give rise to Gabor systems failing to form frames. Our constructions depend only on the parity of the window function g g . Second, for a wide range of finite-dimensional function spaces V V we show that V V contains a function g g such that a lattice of high density fails to generate a Gabor frame. In particular, we prove that the Gröchenig–Lyubarskii theorem is sharp in the finite-dimensional space of polynomials with Gaussian weight. More precisely, for every N ∈ ℕ N\in{\mathbb{N}} and every α , β > 0 \alpha,\beta>0 satisfying α ​ β = 1 N + 1 \alpha\beta=\frac{1}{N+1} , we give an explicit algorithm for finding an even or odd polynomial p p of degree at most N N such that 𝒢 ⁡ ( p ⁡ ( x ) ​ e − π ​ x 2 , α ​ ℤ × β ​ ℤ ) \mathcal{G}(p(x)e^{-\pi x^{2}},\alpha\mathbb{Z}\times\beta\mathbb{Z}) does not form a frame. The proofs are constructive, elementary, and based on linear algebra. 1. Introduction The fundamental problem in Gabor analysis is to determine what assumptions should be imposed on the function g and discrete sets \Lambda,\Gamma to ensure that the Gabor system \mathcal{G}(g,\Lambda\times\Gamma)=\left\{g(x-\lambda)e^{2\pi i\gamma x},\quad\lambda\in\Lambda,\,\gamma\in\Gamma\right\} constitutes a frame for L^{2}({\mathbb{R}}). Formulated at this level of generality, the problem is known to be highly challenging even for a fixed function g . A complete answer is currently available only for the Gaussian kernel g(x)=e^{-ax^{2}},a>0 (see [ 20 , 21 ] ), the Cauchy kernel \frac{1}{x+i} (see [ 3 ] ), and the one-sided exponential e^{-x}\chi_{[0,+\infty)}(x) (see [ 2 ] ). Imposing a lattice structure on at least one of the sets \Lambda and \Gamma places the problem in a significantly more accessible setting, where several complete and partial characterizations are available. For a given function g we define its frame set by \mathcal{F}(g)=\left\{(\alpha,\beta)\in{\mathbb{R}}^{2}_{>0}:\quad\mathcal{G}(g,\alpha{\mathbb{Z}}\times\beta{\mathbb{Z}})\text{ is a frame for }L^{2}({\mathbb{R}})\right\}. For example, the frame set has been completely characterized for totally positive functions, rational functions with poles in a common half-plane, indicator functions of intervals, one-sided exponentially decaying functions, and the first Haar function, see [ 2 , 4 , 5 , 6 , 11 , 14 , 17 ] . However, for many important generators, e.g., Hermite functions, B-splines, ratios of exponential polynomials, the structure of the frame sets is more complicated, and only some sufficient conditions are available, see [ 9 , 13 , 23 ] . These conditions allow us to identify some region \mathcal{S} which is contained entirely inside the frame set \mathcal{F}(g). In particular, for the N -th Hermite function or more generally, for polynomials of degree at most N multiplied by a Gaussian weight, in [ 9 ] it was proved that one can take \mathcal{S}:=\left\{(\alpha,\beta)\in{\mathbb{R}}^{2}_{>0}:\alpha\beta<\frac{1}{N+1}\right\}. On the other hand, for every function g the trivial necessary condition for \mathcal{G}(g,\alpha{\mathbb{Z}}\times\beta{\mathbb{Z}}) to form a frame for L^{2}({\mathbb{R}}) is \alpha\beta<1. Between these two regions, frame properties of Gabor systems have been extensively studied for many families of generators, see e.g. [ 7 , 8 , 15 , 19 ] . In particular, there is a series of results that indicate obstructions for lattices to generate a Gabor system possessing the frame property for Hermite functions and B-splines. For instance, using properties of the Zibulski–Zeevi matrix representation, in [ 19 ] it was proved that for every odd g satisfying mild decay assumptions and m\in{\mathbb{N}}, (1.1) \text{if }\alpha\beta=\frac{m}{m+1}\quad\text{then}\quad\mathcal{G}(g,\alpha{\mathbb{Z}}\times\beta{\mathbb{Z}})\,\,\text{does not form a frame for}\,\,L^{2}({\mathbb{R}}). In the present note, we continue our investigation of the obstructions to the frame property for the Gabor systems \mathcal{G}(g,\alpha{\mathbb{Z}}\times\beta{\mathbb{Z}}) and \mathcal{G}(g,\Lambda\times{\mathbb{Z}}) generated by regular or semi-regular sets. Our approach relies on the classical correspondence (see [ 16 ] ) between Gabor frames and the uniqueness property in shift‑invariant spaces. Our results fall into two categories. First, for a fixed even or odd generator g satisfying mild decay assumptions, we construct explicit examples of the sets \Lambda (including, in particular, certain lattices) of large density for which the Gabor system \mathcal{G}(g,\Lambda\times{\mathbb{Z}}) does not form a frame, see Theorem 3.1 and Corollary 3.2 below. Second, we show that for a collection of windows g_{1},\ldots,g_{N} satisfying suitable decay, linear-independence, and parity assumptions, one can construct periodic sets which are not uniqueness sets for V^{\infty}(g_{1},\ldots,g_{N}) and have density 2N in the odd case and 2N-1 in the even case, see Theorem 3.5. Moreover, these sets already fail to be uniqueness sets for V^{\infty}(g) for a suitable non-trivial function g\in\operatorname{Span}(g_{1},\ldots,g_{N}) , see Corollary 3.6. In addition, in Corollary 3.7 , we show that for every N\in{\mathbb{N}} and every \alpha,\beta,a\in{\mathbb{R}}_{>0} satisfying \alpha\beta=\frac{1}{N+1} , there exists a polynomial p=p(\alpha,\beta,a) such that the Gabor system \mathcal{G}(g,\alpha{\mathbb{Z}}\times\beta{\mathbb{Z}}) does not form a frame for L^{2}({\mathbb{R}}) , where g(x)=p(x)e^{-ax^{2}}. Furthermore, the coefficients of p can be computed explicitly, or approximated numerically with arbitrary precision, by solving a finite system of linear equations. In addition, the polynomial p may be chosen to be even or odd according to the parity of N . In particular, this shows that the boundary for the frame region obtained in [ 9 ] is sharp in the class of polynomials of degree at most N multiplied by a Gaussian. The paper is organized as follows. In Sec. 2 we recall the definition and some properties of shift-invariant spaces. In Sec. 3 we present our main results – Theorems 3.1 , 3.5 and Corollaries 3.2 , 3.6 , 3.7 . The proofs of these results are given in Sec. 4 and Sec. 5. 2. Preliminaries Throughout this paper we will assume that the generator function belongs to the continuous Wiener amalgam space W_{0} which consists of continuous functions g:{\mathbb{R}}\to{\mathbb{C}}, satisfying (2.1) \|g\|_{W_{0}}:=\sum\limits_{k\in{\mathbb{Z}}}\|g\|_{L^{\infty}(k,k+1)}<\infty. Given g\in W_{0} and 1\leq p\leq\infty , the shift-invariant space V^{p}(g) is defined by V^{p}(g)=\{f(x)=\sum\limits_{n\in{\mathbb{Z}}}c_{n}g(x-n),\quad\{c_{n}\}\in l^{p}({\mathbb{Z}})\}. In what follows, we always assume that the function g has stable {\mathbb{Z}} -shifts, i.e. for every p\in[1,\infty] there exist positive constants A,B such that (2.2) A\|\{c_{n}\}\|_{l^{p}}\leq\|f\|_{L^{p}({\mathbb{R}})}\leq B\|\{c_{n}\}\|_{l^{p}},\quad\text{for every }f\in V^{p}(g). This fundamental condition ensures that V^{p}(g) is a closed subset of L^{p}({\mathbb{R}}). Let us briefly recall some facts concerning the stability of integer shifts, for more information and some generalizations we refer the reader to [ 10 , 18 , 22 , 24 ] . First, if ( 2.2 ) is satisfied for some p_{0}\in[1,\infty] then it is also satisfied for all p\in[1,\infty] simultaneously, see [ 10 ] . Second, stability of {\mathbb{Z}} -shifts of g is equivalent to the assumption that there is no x\in[0,1) such that (2.3) \hat{g}(x+n)=0,\quad\text{for every }n\in{\mathbb{Z}}, where \hat{g} stands for the Fourier transform of g defined by \hat{g}(t)=\int\limits_{{\mathbb{R}}}g(x)e^{-2\pi ixt}\,dt. A set \Lambda is called a uniqueness set (US) for the space V^{p}(g) if f(\lambda)=0\,\,\text{for every }\lambda\in\Lambda\quad\text{implies }\quad f\equiv 0. To prove that a given Gabor system does not form a frame, we use the following connection between Gabor systems and shift-invariant spaces, see e.g., [ 12 ] . Lemma 2.1 . Assume that g\in W_{0} has stable integer shifts. Then if for some x\in[0,1) the set \Lambda+x is not a US for

Read the complete text