Extreme and exposed points of shift-invariant spaces generated by Gaussian kernel and hyperbolic secant

Markus Valås Hagen, Alexander Ulanovskii, Denis Zelent, Ilya Zlotnikov · 2026

Read Extreme and exposed points of shift-invariant spaces generated by Gaussian kernel and hyperbolic secant by Markus Valås Hagen, Alexander Ulanovskii, Denis Zelent, 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-2603-04268.txt.

Introduction

For the Gaussian case, Hagen and colleagues identify which points of the L1 unit ball are extreme and which are exposed. The hyperbolic secant case, a quasi shift invariant space, receives the same treatment. Both settings are thus fully characterized, with the Gaussian and hyperbolic secant spaces treated side by side. The results settle the boundary points of each ball.

Catalog record

CREATORMarkus Valås Hagen, Alexander Ulanovskii, Denis Zelent, Ilya Zlotnikov
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE40409
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.04268

Browse the full library · About the Tower of Babel

Full text

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

Extreme and exposed points of shift-invariant spaces generated by Gaussian kernel and hyperbolic secant Thanks: Hagen, Zelent, and Zlotnikov were supported by Grant 334466 of the Research Council of Norway. Markus Valås Hagen, Alexander Ulanovskii, Denis Zelent, Ilya Zlotnikov Abstract . We characterize the extreme and exposed points of the unit ball (with respect to the L 1 L^{1} -norm) in the shift-invariant space generated by the Gaussian function, as well as in the quasi shift-invariant space generated by the hyperbolic secant. 1. Introduction and Main Results 1.1. Extreme and exposed points Let X be a Banach space and let C\subset X be a closed convex set. An element f\in C is called an extreme point of C if it does not lie in the interior of any nontrivial line segment contained in C ; that is, there do not exist distinct points f_{1},f_{2}\in C and a scalar 0<\alpha<1 such that f=\alpha f_{1}+(1-\alpha)f_{2} . Among the extreme points, an element f\in C is said to be an exposed point of C if there exists a continuous linear functional \varphi\in X^{\ast} such that \varphi(f)>\varphi(g)\quad\text{for all }g\in C\setminus\{f\}. Equivalently, \varphi attains its maximum over C uniquely at f . Let B(X)=\{f\in X:\|f\|\leq 1\},\qquad S(X)=\{f\in X:\|f\|=1\} denote the unit ball and the unit sphere of X . We denote the sets of extreme and exposed points of B(X) by {\rm Ext}(X) and \,{\rm Exp}(X), respectively. Note that \,{\rm Exp}(X)\subset{\rm Ext}(X)\subset S(X) . The sets {\rm Ext}(X) and \,{\rm Exp}(X) play a fundamental role in the geometry of the underlying space through the classical Krein–Milman theorem. They arise naturally in a wide range of problems, including the characterization of linear isometries of function spaces [ 6 ] and applications in prediction theory [ 4 ] . These sets have been explored in numerous function spaces, with analytic function spaces receiving particularly sustained attention. The most significant and interesting cases arise for function spaces equipped with the L^{p} -norm when p=1 or p=\infty (see also Remark 1.1 below). In these settings, the extreme and exposed points of the unit ball have been extensively investigated, especially for the classical Hardy spaces and their variants [ 1 , 5 , 11 , 17 ] , various polynomial spaces [ 8 , 9 , 10 , 21 ] , and certain Paley–Wiener spaces [ 8 , 24 ] . For a broader perspective, we refer the reader to the survey [ 12 ] , which also highlights several interesting open problems. This paper presents, to the best of our knowledge, the first study of the extreme and exposed points of the unit ball in shift-invariant and quasi shift-invariant function spaces. We focus on the classical generators—the Gaussian kernel and the hyperbolic secant—so that the resulting spaces consist of entire or meromorphic functions on the complex plane \mathbb{C} . These spaces also play an important role in modern sampling theory and in the theory of Gabor frames. From the perspective of these theories, they exhibit very similar properties due to the relation between the Zak transforms of the Gaussian and hyperbolic secant functions, see [ 15 , 19 , 23 ] . However, as we demonstrate below, their properties differ when examined from the standpoint of Banach space geometry. 1.2. Shift-invariant spaces and quasi shift-invariant spaces Let 1\leq p\leq\infty , and let \Gamma\subset{\mathbb{R}} be a separated set, i.e., \inf\limits_{\gamma,\gamma^{\prime}\in\Gamma,\gamma\neq\gamma^{\prime}}|\gamma-\gamma^{\prime}|>0. The quasi shift-invariant space V_{\Gamma}^{p}(g) generated by a function g consists of the functions f(x)=\sum\limits_{\gamma\in\Gamma}c_{\gamma}g(x-\gamma),\quad\{c_{\gamma}\}\in l^{p}(\Gamma). In what follows, we denote by c_{\gamma}(f) the coefficient c_{\gamma} in the representation above. Following tradition, we assume throughout that the set of \Gamma ‑shifts is relatively dense; that is, there exists an R>0 such that every interval of length R contains at least one point of \Gamma . This assumption is not essential for the subsequent analysis. In the particular case \Gamma={\mathbb{Z}} , the space V^{p}(g):=V^{p}_{{\mathbb{Z}}}(g) is called shift-invariant. The space V_{\Gamma}^{p}(g) is well-defined whenever g belongs to the Wiener amalgam space W , i.e., (1.1) \sum\limits_{k\in{\mathbb{Z}}}\|g\|_{L^{\infty}[k,k+1]}<\infty, since then \|f\|_{L^{p}({\mathbb{R}})} is finite for every f\in V^{p}_{\Gamma}(g) , see e.g. [ 25 , Lemma 8.1] . As usual, we equip the spaces V^{p}_{\Gamma} with the standard L^{p} -norm. The quasi shift-invariant spaces considered below satisfy the following condition: for every p\in[1,\infty] there exist positive constants K_{1},K_{2} such that (1.2) K_{1}\|c\|_{l^{p}}\leq\|f\|_{L^{p}({\mathbb{R}})}\leq K_{2}\|c\|_{l^{p}},\quad\text{ for every $f\in V^{p}_{\Gamma}(g).$} This property shows that V^{p}_{\Gamma}(g) is a closed subspace of L^{p}({\mathbb{R}}) . Note that when \Gamma={\mathbb{Z}} and f\in W , the norm equivalence ( 1.2 ) holds if and only if the Fourier transform \hat{g} does not completely vanish on any arithmetic progression of step 1 , see [ 20 ] . For the case of general separated sets \Gamma , we refer the reader to [ 26 ] . Like the Paley–Wiener spaces, (quasi) shift-invariant spaces play a fundamental role in approximation theory. In particular, the sampling and interpolation properties of the spaces V^{p}_{\Gamma}(g) are closely connected to the frame properties of the Gabor systems generated by g , see, e.g., [ 2 , 14 , 15 , 16 , 23 , 25 ] . Important examples arise when g is the Gaussian kernel or the hyperbolic secant: (1.3) G_{a}(x):=e^{-ax^{2}},\quad H_{a}(x)=\frac{1}{e^{ax}+{e^{-ax}}}, where a>0. It is well-known that the spaces V^{p}(G_{a}) and V^{p}_{\Gamma}(H_{a}) satisfy ( 1.2 ), see e.g. [ 16 ] , [ 25 ] , [ 26 ] . It is easy to check that every element of V^{p}(G_{a}) admits analytic extension to an entire function, while every element of V^{p}_{\Gamma}(H_{a}) admits analytic extension to a meromorphic function. Remark 1.1 . For the spaces V^{p}(G_{a}) and V^{p}_{\Gamma}(H_{a}) with 1<p<\infty , the set of extreme points of the unit ball coincides with the unit sphere. This follows directly from the classical equality case in Minkowski’s inequality: \|f+g\|_{L^{p}({\mathbb{R}})}=\|f\|_{L^{p}({\mathbb{R}})}+\|g\|_{L^{p}({\mathbb{R}})} iff f=ag for some a\geq 0 . The present paper examines the geometry of the unit balls of these spaces in the case p=1. 1.3. Main results Theorem 1.2 . (Gaussian kernel) (i) The set {\rm Ext}(V^{1}(G_{a})) consists of those functions f\in V^{1}(G_{a}) satisfying \|f\|_{1}=1 and such that (1.4) \text{There is no point $\lambda\in\mathbb{C}$ with $0<{\rm Im\,}\lambda<\pi/a$ for which }f(\lambda)=f(\bar{\lambda})=0. (ii) The set \,{\rm Exp}(V^{1}(G_{a})) consists of those functions f\in{\rm Ext}(V^{1}(G_{a})) for which the following two conditions hold: (1.5) \text{There is no point }\lambda\in{\mathbb{R}}\quad\text{such that}\quad f(\lambda)=f^{\prime}(\lambda)=0; (1.6) \int_{\mathbb{R}}e^{2ax}|f(x)|\,dx=\int_{\mathbb{R}}e^{-2ax}|f(x)|\,dx=\infty. Remark 1.3 . Similar conditions to ( 1.4 ) and ( 1.5 ) also arise in the setting of polynomials and Paley–Wiener spaces ; see e.g., [ 8 , Theorems 1 and 3] . Remark 1.4 . (See Lemma 3.3 ) Let f\in V^{1}(G_{a}). Condition ( 1.6 ) is equivalent to requiring that neither sequence c_{n}(f)e^{2an} nor c_{n}{(f)}e^{-2an} belongs to l^{1}({\mathbb{Z}}) , i.e. \sum_{n\in{\mathbb{Z}}}e^{2an}|c_{n}(f)|=\sum_{n\in{\mathbb{Z}}}e^{-2an}|c_{n}(f)|=\infty. For the hyperbolic secant, we obtain the description of {\rm Ext}(V^{1}_{\Gamma}(H_{a})) and \,{\rm Exp}(V^{1}_{\Gamma}(H_{a})) for general separated sets of translates \Gamma. Theorem 1.5 . (Hyperbolic secant generator) Let \Gamma\subset{\mathbb{R}} be a separated relatively dense set. (i) The set {\rm Ext}(V_{\Gamma}^{1}(H_{a})) consists of those functions f\in V_{\Gamma}^{1}(H_{a}) satisfying \|f\|_{1}=1 , condition ( 1.4 ) and the condition (1.7) c_{\gamma}(f)\neq 0,\quad\text{for all \,}\gamma\in\Gamma. (ii) The set \,{\rm Exp}(V^{1}_{\Gamma}(H_{a})) consists of those functions f\in{\rm Ext}(V^{1}_{\Gamma}(H_{a})) satisfying conditions ( 1.5 ) and ( 1.6 ). Remark 1.6 . (See Corollary 4.4 ) For

Read the complete text