Limit Theorems for the Volume of Random Projections and Sections of ℓ p N \ell_{p}^{N} -balls
Joscha Prochno, Christoph Thäle, Philipp Tuchel · 2024
Read Limit Theorems for the Volume of Random Projections and Sections of ℓ p N \ell_{p}^{N} -balls by Joscha Prochno, Christoph Thäle, Philipp Tuchel (2024) 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-2412-16054.txt.
Introduction
Joscha Prochno, Christoph Thäle, and Philipp Tuchel study random slices of high-dimensional balls in this 2024 paper. The balls are defined by a general power norm and grow in dimension. A random flat subspace of fixed size cuts or projects the ball, and the team asks how the resulting volume behaves.
They prove a full set of limit laws: typical fluctuations follow a bell curve, while rarer swings follow moderate and large deviation rules. The work covers plain volume plus related measures such as internal and dual volumes. It extends earlier results that handled only the cube or only small cuts. Together, the theorems give a complete picture of these random shapes.
Catalog record
| CREATOR | Joscha Prochno, Christoph Thäle, Philipp Tuchel |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 111 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-16054
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Limit Theorems for the Volume of Random Projections and Sections of ℓ p N \ell_{p}^{N} -balls Joscha Prochno, Christoph Thäle, Philipp Tuchel Abstract Let 𝔹 p N \mathbb{B}_{p}^{N} be the N N -dimensional unit ball corresponding to the ℓ p \ell_{p} -norm. For each N ∈ ℕ N\in\mathbb{N} we sample a uniform random subspace E N E_{N} of fixed dimension m ∈ ℕ m\in\mathbb{N} and consider the volume of 𝔹 p N \mathbb{B}_{p}^{N} projected onto E N E_{N} or intersected with E N E_{N} . We also consider geometric quantities other than the volume such as the intrinsic volumes or the dual volumes. In this setting we prove central limit theorems, moderate deviation principles, and large deviation principles as N → ∞ N\to\infty . Our results provide a complete asymptotic picture. In particular, they generalize and complement a result of Paouris, Pivovarov, and Zinn [A central limit theorem for projections of the cube, Probab. Theory Related Fields. 159 (2014), 701-719] and another result of Adamczak, Pivovarov, and Simanjuntak [Limit theorems for the volumes of small codimensional random sections of ℓ p n \ell_{p}^{n} -balls, Ann. Probab. 52 (2024), 93-126]. Keywords . Asymptotic theory of convex bodies, central limit theorem, high-dimensional probability, functional limit theorem, large deviation principle, ℓ p N \ell_{p}^{N} -ball, moderate deviation principle, random projection, random section, stochastic geometry MSC . Primary 52A23, 60F05, 60F10; Secondary 46B09, 52A22, 60D05. 1 Introduction The asymptotic theory of convex bodies plays a major role in the fields of Asymptotic Geometric Analysis and High-Dimensional Probability, two rather young branches of mathematics having their origin in the Local Theory of Banach spaces and Probability in Banach Spaces; we refer to [ 7 ] and [ 77 ] . The two fields and, in particular, the methods used and developed during the last decades are closely linked to convex and discrete geometry, functional analysis, and probability theory, and have a variety of applications in, for instance, compressed sensing and image processing [ 13 , 25 ] , information-based complexity [ 31 , 32 ] , and machine learning [ 22 , 62 , 63 ] , to mention just a few. The probabilistic perspective on high-dimensional convex bodies and structures is an extremely powerful one and has led to several major breakthrough results. We would like to mention here the central limit theorem for convex bodies obtained by Klartag [ 53 ] or Eldan’s stochastic localization method [ 21 ] , which eventually led to the spectacular solution of the slicing conjecture in [ 55 ] (see also [ 15 , 68 , 54 , 58 ] for previous works in this direction). In the last decades a number of weak limit theorems related to convex bodies and high-dimensional random geometric structures and quantities have appeared and we refer the reader to the following non-exhaustive list and the references cited therein [ 5 , 10 , 34 , 35 , 36 , 41 , 69 , 71 , 72 , 75 , 76 ] . Less than a decade ago, the central limit perspective has been complemented by moderate and large deviation principles through the works of Gantert, Kim, and Ramanan [ 28 ] and Kabluchko, Prochno, and Thäle [ 44 ] . While universality results are mathematically elegant and also quite powerful, from this perspective, high-dimensional convex bodies cannot be distinguished, for instance, by their lower-dimensional random projections. However, the latter play a fundamental role, e.g., in dimension reduction. Large and moderate deviation principles however are known to be typically non-universal and distribution-dependent in that they are parametric in speed and/or rate function. In this sense, it is reasonable to expect that the large and moderate deviation behavior of, for instance, a random projection of a convex body, depends in a more subtle way on the geometry of the underlying body. The goal to precisely identify how a random projection can encode distinct distributional information about the original vector, triggered an intensive research activity and a whole variety of large and moderate deviation principles have been obtained in the frameworks of Asymptotic Geometric Analysis and High-Dimensional Probability since [ 4 , 6 , 26 , 27 , 43 , 47 , 48 , 51 , 52 , 59 , 60 ] . Moreover, new tools, such as maximum entropy considerations and elements from log-potential theory, emerged from the theory of large deviations and the related field of statistical mechanics and turned out to be quite powerful, leading to a number of breakthrough results [ 11 , 17 , 37 , 38 , 39 , 40 , 42 , 50 ] . For the current state of the art we refer to the recent survey [ 67 ] on the large and moderate deviations approach in geometric functional analysis. The motivation for the present paper stems at first from a result obtained by Paouris, Pivovarov, and Zinn who proved the following central limit theorem in [ 66 ] . For N\in\mathbb{N} let {\mathbb{B}}_{\infty}^{N}=[-1,1]^{N} be the unit N -dimensional cube and assume that m\in\mathbb{N} with m\leq N . Denote by \mathbb{G}_{m,N} the Grassmannian manifold of m -dimensional linear subspaces of \mathbb{R}^{N} , which is equipped with the Haar probability measure \mu_{m,N} , and suppose that (E_{N})_{N\geq m} is a sequence of independent random subspaces such that E_{N}\in\mathbb{G}_{m,N} has distribution \mu_{m,N} . Define for each N\in\mathbb{N} a random variable Z_{N}:=\mathrm{vol}_{m}\big({\mathbb{B}}_{\infty}^{N}\big|E_{N}\big), where {\mathbb{B}}_{\infty}^{N}\big|E_{N} denotes the orthogonal projection of B_{\infty}^{N} onto the linear subspace E_{N} . Then \frac{Z_{N}-\mathbb{E}[Z_{N}]}{\sqrt{\mathbb{V}[Z_{N}]}}\xrightarrow[N\to\infty]{\textup{d}}\mathcal{N}(0,1), where \stackrel{{\scriptstyle\textup{d}}}{{\rightarrow}} denotes convergence in distribution. At its core, the proof is based on the Gaussian projection formula \mathrm{vol}_{m}(G{\mathbb{B}}_{\infty}^{N})=\det(GG^{*})^{\frac{1}{2}}\mathrm{vol}_{m}\big({\mathbb{B}}_{\infty}^{N}\big|E_{N}\big) with a standard Gaussian matrix G\in\mathbb{R}^{m\times N} having independent columns g_{1},\dots,g_{N} ( G^{*} denotes the transpose of G ) and the fact that GB_{\infty}^{N} is a random zonotope (i.e., the Minkowski sum of the random line segments [-g_{i},g_{i}] ). The latter allows a representation of the volume of the Gaussian projection essentially in terms of a U -statistic (or a special case of Minkowski’s theorem on mixed volumes of convex sets, to put it in convex geometry parlance), which assures that one can apply Vitale’s central limit theorem for Minkowski sums of random convex sets [ 78 ] ; one should note, however, that both the Gaussian projection as well as the determinantal part contribute to the asymptotic normality, which requires a careful and delicate analysis (we refer to [ 66 ] for more information). Another motivation for this paper comes from the work of Adamczak, Pivovarov, and Simanjuntak [ 2 ] . In contrast to what has been discussed before, these authors study for general \ell_{p} -balls in \mathbb{R}^{N} with 0<p\leq\infty limit theorems for volumes of random hyperplane sections; see Section 2 for formal definitions. More precisely, if {\mathbb{B}}_{p}^{N} denotes the \ell_{p} -ball in \mathbb{R}^{N} and if (H_{N})_{N\geq 1} is a sequence of independent random elements of the Grassmannian \mathbb{G}_{N-1,N} of hyperplanes in \mathbb{R}^{N} , they show that there is a sequence (c_{N})_{N\geq 1} such that N^{3/2}\Bigg(\frac{\mathrm{vol}_{N-1}({\mathbb{B}}_{p}^{N}\cap H_{N})}{\mathrm{vol}_{N-1}({\mathbb{B}}_{p}^{N-1})}-c_{N}\Bigg)\xrightarrow[N\to\infty]{\textup{d}}\xi where \xi\sim\mathcal{N}(0,\sigma_{p}^{2}) and both, (c_{N})_{N\geq 1} as well as \sigma_{p}^{2} , are given explicitly (see [ 2 , Theorem 3.4] ). A similar result is also obtained for the volumes of the sections with a random subspace of a fixed codimension m\in\mathbb{N}