On the dual form of `low \(M^{*}\) estimate' in the quasi-convex case (Q1780506): Difference between revisions

The paper under review contains a proof of the following Theorem. Suppose that \(K \subset \mathbb{R}^n\) is a \(p\)-convex body for some \(0 < p < 1\) (that is, if \(x, y \in K\) and \(\lambda^p + \mu^p \leq 1\), then also \(\lambda x + \mu y \in K\)). Let \(M_K\) stand for the average of the gauge function \(\| x \| _K = \inf \{ t > 0: x \in t K \}\) on the unit sphere \(S^{n-1}\) in \(\mathbb{R}^n\). Fix \(0 < \lambda < 1\) and let \(E\) be a random \(\lfloor (1 - \lambda) n \rfloor\)-dimensional subspace in \(\mathbb{R}^n\). Then, with high probability, \[ c_p \frac{\lambda^{\frac{1}{p} - \frac{1}{2}}}{M_K} B_E \subset \text{Proj}_E(K), \] where \(B_E\) is the Euclidean unit ball in the subspace \(E\), \(\text{Proj}_E\) is the orthogonal projection onto \(E\) in \(\mathbb{R}^n\), and \(c_p\) is a constant that depends only on \(p\). The result is a \(p\)-convex counterpart to the (dual) convex low \(M^*\)-estimate, a useful inequality in high-dimensional convex geometry due to Milman that was further developed by Pajor, Tomczak--Jaegermann and Gordon (see references in the paper under review). The theorem improves upon a previous result [\textit{A.~E.\ Litvak, V.~D.\ Milman} and \textit{A.~Pajor}, Proc.\ Am.\ Math.\ Soc.\ 127, No.~5, 1499--1507 (1999; Zbl 0948.52005)] in which the dependence on \(\lambda\) is weaker. The proof presented here is very short: using a result of Gordon and Kalton, one may reduce matters to the convex case.