Id: arxiv/2607.23017 Url: https://arxiv.org/abs/2607.23017 Tags: math.PR, cs.DM, cs.DS Score: 0 Title: Level-set entropy and sparse randomized embeddings Domain: math.PR Source: arxiv Authors: Konstantin Tikhomirov Reviews: 0 Subject: math.PR Category: research Comments: 0 Disputed: false Division: math Mentions: 0 Discussed: 0 Published: 2026-07-25 Page: https://delphi.science/r/arxiv/2607.23017 Let $Π$ be a $k\times n$ sparse random matrix. For a fixed $r$-dimensional subspace $V\subset{\mathbb R}^n$, let $U_V:{\mathbb R}^r\to{\mathbb R}^n$ denote an isometry from ${\mathbb R}^r$ onto $V$. The product $ΠU_V$ is a central model in randomized dimension reduction and has been studied primarily through trace and Gaussian comparison inequalities. In this work, we develop an approach to the spectral norm of the matrix product $ΠU_V$, based on entropy estimates for level sets of vectors $x\in V$. Combining the method with existing estimates, we show the following. Assume that \[ k\ge…