Constructive Decompositions of the Identity for Functional John Ellipsoids
Fernanda Moreira Baêta · 2025
Read Constructive Decompositions of the Identity for Functional John Ellipsoids by Fernanda Moreira Baêta (2025) 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-2504-01631.txt.
Introduction
Baêta studies functional John ellipsoids and how to build a decomposition of the identity for them, in a mathematics paper from 2025. Functional ellipsoids, in the sense set out by Ivanov and Naszódi, extend the classic ellipsoid tools of convex geometry to functions. Fritz John's famous theorem gives such a decomposition for ordinary ellipsoids, with weights and contact points that sum to the identity. Baêta asks how to construct the same kind of decomposition in the functional setting, step by step rather than by a pure existence claim.
Catalog record
| CREATOR | Fernanda Moreira Baêta |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 76 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-01631
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Constructive Decompositions of the Identity for Functional John Ellipsoids Fernanda Moreira Baêta Abstract . We consider functional ellipsoids in the sense defined by Ivanov and Naszódi [ 8 ] and we study the problem of constructing a decomposition of the identity similar to the one given by Fritz John in his fundamental theorem. 1. Introduction and Main Results In 1948, Fritz John established that every convex body K\subset\mathbb{R}^{n} contains a largest volume ellipsoid. When this ellipsoid is the n -dimensional unit Euclidean ball B^{n} , the body K is said to be in John position. John’s theorem further guarantees the existence of a finite set of points \{\xi_{1},\dots,\xi_{m}\}\subset S^{n-1}\cap\partial K , positive weights \{c_{1},\dots,c_{m}\} , and a scalar \lambda\neq 0 such that the following conditions hold (1) \displaystyle\sum_{i=1}^{m}c_{i}\xi_{i}\otimes\xi_{i}=\lambda\Id\qquad\mbox{and}\qquad\sum_{i=1}^{m}c_{i}\xi_{i}=0. Here v\otimes w denotes the rank-one matrix vw^{T} , \Id is the n\times n identity matrix, S^{n-1} is the unit Euclidean sphere, and \partial K is the boundary of K (see [ 14 , Application 4, pag. 199 - 200]). This necessary condition also holds for the dual case where the unit Euclidean ball is the ellipsoid with minimum volume containing K . We say in this case that K is in Löwner position. The decomposition ( 1 ) is often referred to as the decomposition of the identity. As shown by Ball [ 4 ] , the existence of a measure \mu_{K} , supported on S^{n-1}\cap\partial K , satisfying (2) \displaystyle\int_{S^{n-1}}(\xi\otimes\xi)d\mu_{K}=\lambda\Id\qquad\mbox{and}\qquad\int_{S^{n-1}}\xi d\mu_{K}=0, for some \lambda\neq 0 , ensures that K is in John position if B^{n}\subseteq K , or in Löwner position if K\subseteq B^{n} . A measure satisfying ( 2 ) is called isotropic and centered, respectively. The John and Löwner ellipsoids form a cornerstones of modern convex geometry, and many problems has been solved using properties of these objects. The relationship between isotropic measures and extremal positions has been widely studied, including by [ 9 , 10 , 11 ] . Extensions to related minimization problems appear in [ 5 , 6 , 12 , 16 , 17 ] . More recently, the theory of ellipsoids has been extended to the space of log-concave functions. In 2018, Alonso-Gutiérrez, Gonzales Merino, Jiménez and Villa [ 1 ] extended the notion of the John ellipsoid to log-concave functions. A function \varphi:\mathbb{R}^{n}\rightarrow(-\infty,+\infty] is convex if, for all x,y\in\mathbb{R}^{n} and \lambda\in[0,1] , \varphi(\lambda x+(1-\lambda)y)\leq\lambda\varphi(x)+(1-\lambda)\varphi(y). When h=e^{-\varphi} for a convex function \varphi , h is called a log-concave function . Given an integrable log-concave function h:\mathbb{R}^{n}\to[0,\infty) , they defined its John ellipsoid as follows: For a fixed constant \beta\in(0,\|h\|_{\infty}) , consider the superlevel set \{x\in\mathbb{R}^{n}:h(x)\geq\beta\} , which is a bounded convex set with non-empty interior. For each level \beta>0 , let \mathcal{E} be the maximal-volume ellipsoid contained within this superlevel set. They proved the existence of a unique height \beta_{0}\in[0,\|h\|_{\infty}] that maximizes \beta_{0}\A_{n}(\mathcal{E}) , where \A_{n} denotes the Lebesgue measure. The John ellipsoid of h is then defined as the function \mathcal{E}^{\beta_{0}}(x)=\beta_{0}{1}_{\mathcal{E}}(x) , where {1}_{\mathcal{E}} is the indicator function of \mathcal{E} . In 2019, Li, Schütt, and Werner [ 15 ] introduced the dual notion of the Löwner ellipsoid for log-concave functions. They showed that for any non-degenerate, integrable log-concave function h , there exists a unique pair (A_{0},t_{0}) , where A_{0} is an invertible affine transformation and t_{0}\in\mathbb{R} , such that \int_{\mathbb{R}^{n}}e^{-|A_{0}x|_{2}+t_{0}}\,dx=\min\left\{\int_{\mathbb{R}^{n}}e^{-|Ax|_{2}+t}\,dx:e^{-|Ax|_{2}+t}\geq h(x)\right\}. Here, |x|_{2} denotes the Euclidean norm of x\in\mathbb{R}^{n} . The function e^{-|A_{0}x|_{2}+t_{0}} is called the Löwner function of h . We say that h:\mathbb{R}^{n}\rightarrow\mathbb{R} is upper semicontinuous if \limsup_{k\rightarrow+\infty}h(x_{k})\leq h(x), whenever x_{k}\rightarrow x as k\rightarrow+\infty . A log-concave function h on \mathbb{R}^{n} is said to be proper if h is upper semicontinuous and has finite positive integral. We will say that a function f:\mathbb{R}^{n}\rightarrow\mathbb{R} is below a function g:\mathbb{R}^{n}\rightarrow\mathbb{R} if f(x)\leq g(x) for all x\in\mathbb{R}^{n} . Recently, in 2021, Ivanov and Naszódi [ 8 ] also extended the notion of the John ellipsoid to the setting of logarithmically concave functions. Unlike the first ones, they defined a class of functions on \mathbb{R}^{n} indexed by s>0 . First they fix a log-concave function h:\mathbb{R}^{n}\rightarrow[0,\infty) and a parameter s>0 . Later, they prove that there exists (and is unique within the set of log-concave functions) a log-concave function with the largest integral under the condition that it is pointwise less than or equal to h^{1/s} . This function is called the John s -function of h . In [ 8 , Theorem 6.1], it is shown that as s\rightarrow 0 , the John s -functions converge to characteristic functions of ellipsoids, thereby establishing a relationship between the first [ 1 ] and second approach [ 8 ] . Furthermore, the authors study the behavior of the John s -functions as s\to\infty , demonstrating that the limit may only be a Gaussian density, which is not necessarily unique. Ivanov and Tsiutsiurupa [ 13 ] , in 2021, studied the dual problem of the John s -function of a log-concave function defined in [ 8 ] and introduced the Löwner s -function. They combined ideas from [ 15 ] and [ 8 ] as following. For a function \psi:[0,+\infty)\rightarrow(-\infty,+\infty] , they considered the class of functions \alpha e^{-\psi(|A(x-a)|_{2})} , where A is invertible, \alpha>0 , and a\in\mathbb{R}^{n} , as the class of “affine” positions of the function x\mapsto e^{-\psi(|x|_{2})},x\in\mathbb{R}^{n} . Note that these problems are related because the classes of “affine” positions of the characteristic function of the unit ball were considered in [ 1 ] , while the classes of “affine” positions of the function x\mapsto e^{-|x|_{2}},x\in\mathbb{R}^{n} , were studied in [ 15 ] . For a function \psi:[0,+\infty)\rightarrow(-\infty,+\infty] such that e^{-\psi(|t|)},t\in\mathbb{R} , is an upper semicontinuous log-concave function with a finite positive integral, and an upper semicontinuous log-concave function h:\mathbb{R}^{n}\rightarrow[0,+\infty) of finite positive integral, they studied the following optimization problem \displaystyle\int_{\mathbb{R}^{n}}\alpha_{0}e^{-\psi(|A_{0}(x-a_{0})|_{2})}dx=\min\left\{\int_{\mathbb{R}^{n}}\alpha e^{-\psi(|A(x-a)|_{2})}dx:h(x)\leq\alpha e^{-\psi(|A(x-a)|_{2})}\right\}. The height function of the (n+1) -dimensional unit ball B^{n+1}\subset\mathbb{R}^{n+1} , given by \hslash_{B^{n+1}}(x)=\sqrt{1-|x|_{2}^{2}} if x\in B^{n} and 0 otherwise, is a proper log-concave function. The main advantage of the framework presented in [ 8 ] is that it implies a “decomposition of the identity” as in ( 1 ). Namely, Theorem 1.1 ( [ 8 ] , Theorem 5.2) . Let h be a proper log-concave function on \mathbb{R}^{n} and s>0 . Assume \hslash^{s}_{B^{n+1}}\leq h . Then the following conditions are equivalent: (1) The function \hslash^{s}_{B^{n+1}} is the John s -function of h ; (2) There exist points u_{1},\dots,u_{k}\in B^{n}\subset\mathbb{R}^{d} and positive weights c_{1},\dots,c_{k} , such that (a) h(u_{i})=\hslash^{s}_{B^{n+1}}(u_{i}) for all i=1,\dots,k ; (b) \sum_{i=1}^{k}c_{i}u_{i}\otimes u_{i}=\Id ; (c) \sum_{i=1}^{k}c_{i}h(u_{i})^{1/s}h(u_{i})^{1/s}=s ; (d) \sum_{i=1}^{k}c_{i}u_{i}=0. Similar to the decomposition of the identity in the geometric case, the authors of [ 8 ] guarantee the existence of a measure satisfying (a)-(d) in the theorem above, although the proof is not constructive. In [ 2 ] , the authors presented a constructive proof of John’s theorem in the geometric setting, using a simple finite-dimensional minimization problem. In this context, assuming that \hslash^{s}_{B^{n+1}} is the John s -function of h , the purpose of this paper is to present a constructive proof of the necessity part of Theorem 1.1 , using as in [ 2 ] a simple finite dimensional minimization problem. Throughout this paper, we denote by \W_{d}(\mathbb{R}) the vector space of d\times