Forward and reverse entropy power inequalities in convex geometry (Q2406330)

Given a random variable \(X\) with density \(f(x)\) in \(\mathbb R^d\), the Shannon-Boltzmann entropy of \(X\) is defined as \(h(X)=\mathbb E[-\log f(X)]\), and its \textit{entropy power} by \(N(X)=e^{2h(X)/d}\). Then, the entropy power inequality (EPI for short) states that \[ N(X+Y)\geq N(X)+N(Y), \] and it can be seen as an analog of the famous Brunn-Minkowski inequality. As a generalization of the Shannon-Boltzmann entropy comes out the so-called \textit{Rényi entropies}: for \(p\in(0,\infty)\backslash\{1\}\), the \(p\)-Rényi entropy of \(X\) is defined by \[ h_p(X)=\frac{1}{1-p}\log\int_{\mathbb R^d}f(x)^p\,dx. \] By continuity, \(h_1(X)=h(X)\) is the Shannon-Boltzmann entropy. In this survey the authors review several types of EPIs, both for the usual Shannon-Boltzmann entropy and for the more general \(p\)-Rényi entropies. They also discuss reverse EPIs, which involve particular linear transformations. The paper concludes by considering integral and entropic liftings of the classical Blaschke-Santaló inequality. For the entire collection see [Zbl 1377.52002].