Duality for Delsarte’s extremal problem on locally compact Abelian groups

Elena E. Berdysheva, Bálint Farkas, Marcell Gaál, Mita D. Ramabulana, Szilárd Gy. Révész · 2026

Read Duality for Delsarte’s extremal problem on locally compact Abelian groups by Elena E. Berdysheva, Bálint Farkas, Marcell Gaál, Mita D. Ramabulana, Szilárd Gy. Révész (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-2603-18287.txt.

Introduction

Berdysheva and co-authors prove strong duality for Delsarte's extremal problem on locally compact Abelian groups. The problem began in coding theory as a bound on error-correcting code size. It now serves sphere packing, spectral sets, and 1-avoiding sets as well. They widen both the normalization and the objective to cover many past cases at once. A functional analytic proof, with harmonic analysis, unifies earlier duality results.

Catalog record

CREATORElena E. Berdysheva, Bálint Farkas, Marcell Gaál, Mita D. Ramabulana, Szilárd Gy. Révész
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61970
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.18287

Browse the full library · About the Tower of Babel

Full text

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

Duality for Delsarte’s extremal problem on locally compact Abelian groups Elena E. Berdysheva, Bálint Farkas, Marcell Gaál, Mita D. Ramabulana, Szilárd Gy. Révész Abstract . The Delsarte extremal problem for positive definite functions, originally introduced by Delsarte in coding theory to bound the size of error-correcting codes, has since found applications in diverse areas such as sphere packing, Fuglede’s spectral set conjecture, and 1 1 -avoiding sets. Recent developments have established the existence of extremizers in fairly general settings and identified precise linear programming dual formulations, together with strong duality results, in several important cases including finite groups and ℝ d \mathbb{R}^{d} . In this paper, we consider a generalized Delsarte problem on locally compact Abelian groups, providing a natural framework for harmonic analysis. We extend both the normalization and the objective functional to encompass a wide range of previously studied cases, while avoiding restrictive topological assumptions common in the literature. Within this general setting, we derive the corresponding dual problem and prove a strong duality theorem, thereby unifying and extending earlier results. Naturally, our proof uses harmonic analysis, but the key is a functional analytic approach which distinguishes our proof from existing methods. 2020 Mathematics Subject Classification Primary: 43A35. Secondary: 43A25. 1. Introduction The Delsarte extremal problem originates from coding theory, where the maximal possible size of codes of given length and self-correcting at a prescribed extent were estimated via their Krawtchouk polynomial expansions. See [ 19 , 20 ] , or for a survey [ 11 ] . It turned out fast that the approach can be reconfigured to more common setups, where non-negativity of Krawtchuk polynomial expansion coefficients corresponded to non-negativity of the Fourier transform – that is, positive definiteness. For early development of the idea see [ 45 ] [ 39 ] [ 36 ] [ 66 ] , [ 31 ] , [ 30 ] [ 16 ] . The Delsarte extremal problem on positive definite functions for various sets in \mathbb{R}^{d} received much attention because in many instances it provides the best available tool—sometimes even a sharp tool—to access difficult problems. This was seen in case of sphere packings, [ 45 ] [ 66 ] [ 63 ] [ 14 ] [ 17 ] [ 31 ] , and in the case of the Fuglede Conjecture, too [ 44 ] . Let us mention here the Erdős-Moser distance avoiding set conjecture [ 18 , 21 ] , which has seen a long development until it was proved recently in [ 2 ] . The Delsarte scheme has already been extended to locally compact Abelian groups. For example, estimates of the upper density of the translational vectors T in a packing T+\Omega by translates of a given set \Omega are given by the Delsarte bound in LCA groups [ 10 ] . The Delsarte extremal problem can be formulated as an extremal problem on a function class \mathcal{F} . However, its formulation seems to depend on the choice of the function class, and it is far from obvious which class to consider. To be more precise, in the classical problem of sphere packing in \mathbb{R}^{d} , Gorbachev in [ 31 ] worked with band-limited functions, Viazovska in [ 63 ] , and again Viazovska and coauthors in [ 17 ] used classes of functions which, along with their Fourier transform, decay sufficiently fast (with a polynomial speed), whereas Cohn and Elkies [ 16 ] and Cohn, de Laat, Salmon [ 15 ] considered the Schwartz class, etc.. The non-trivial equivalence of these, a priori different, extremal problems is worked out in [ 10 ] relying on useful personal communications from D. Gorbachev. The general definition can be formulated as follows. Definition 1.1 . Let G be a locally compact Abelian group 1 1 1 Including the Hausdorff property. with neutral element 0 and Haar measure \lambda . Denote the family of continuous positive definite functions by {\mathcal{D}}={\mathcal{D}}(G) . Let \mathcal{F} be a function class, forming a subspace of C(G;\mathbb{R}) (real-valued continuous functions on G ) and let \Omega\subseteq G be a symmetric set with 0\in\mbox{\rm int\,}{\Omega} , having compact closure. Then the Delsarte constant is (1) D_{G}(\mathcal{F},\Omega):=\sup\left\{\int_{G}f\mathrm{d}\lambda:f\in{\mathcal{D}}\cap\mathcal{F},f(0)=1,f|_{G\setminus\Omega}\leq 0\right\}. In fact, in this work we shall generalize this setup to the respective extremal problems with both the Haar integral (i.e., the goal functional) and the point evaluation at 0 (i.e., normalizing functional) replaced by other almost arbitrary linear functionals. Some normalization for the considered functions is necessary: fully dropping any normalization constraints leaves us with a full convex cone \mathcal{F}\cap{\mathcal{D}} , where the goal functional \rho can become unbounded. However, note that regarding the underlying set \Omega the formulation is quite general. Indeed, we require no topological conditions on \Omega apart from 0\in\mbox{\rm int\,}\Omega , which is necessary, as otherwise the normalization condition on f(0)=1 trivialize the function set to \emptyset . Symmetry is not a restriction either, as we can always consider \Omega\cap(-\Omega) instead of the original (possibly not symmetric) set, given that any positive definite function f\in{\mathcal{D}} , vanishing outside \Omega , vanishes outside -\Omega , too. Finally, compact closure can be relaxed, too, as we will discuss later, but some type of boundedness is necessary, if we want D_{G}(\mathcal{F},\Omega) to remain finite. Obviously, the extremal problem of Delsarte is an extremal problem of infinite dimensional linear programming type, where the linear conditions bounding the feasibility set can be expressed in several ways. One direct, but rarely used variant is to say that positive definiteness is encoded in the usual inequalities \sum_{k=1}^{n}\sum_{\ell=1}^{n}c_{k}\overline{c_{\ell}}f(x_{k}-x_{\ell})\geq 0 with all coefficient sequences (c_{k})_{k=1}^{n}\in\mathbb{C}^{n} , any n\in\mathbb{N} . In fact, here is a hidden difficulty in front of us. We are doing real linear programming duality, therefore we restrict to real-valued positive definite functions, and this class is slightly different from standard positive definite functions. This issue was properly addressed in [ 27 ] , and we will follow that work. A second possibility is to use the non-negativity of the Fourier transform \widehat{f} of f , which is equivalent to positive definiteness whenever \mathcal{F}\subseteq L^{1}(G;\mathbb{R}) , too. Nevertheless, we need to see here that our class restriction gives the same extremal values as other usual class definitions. Now, general linear programming setup requires an image space Z , and then we can say that the conditioning in D_{G}(\mathcal{F},\Omega) consists of two parts, one in X (the inequalities f(x)\leq 0 on points outside \Omega ), and another one (of non-negativity of the Fourier transform) after considering the operator of the Fourier transform T:X\to Z . For duality we then need to consider dual spaces Y=X^{\prime} and W=Z^{\prime} , and determine the (bounded, linear) adjoint operator T^{*}:W\to Y . Then to have the strong duality result, restrictive properties of the adjoint may be needed. In fact, this “standard linear programming approach” was worked out first in [ 4 ] , and in our previous paper [ 9 ] we followed this direction. However, we could carry over the arguments only under some compactness conditions, even if in that case in a great generality (with, e.g., two-sided sign conditions). Another way to look at the duality problem is to say that we do not want to leave the space X , so

Read the complete text