Regular random sections of convex bodies and the random quotient-of-subspace theorem (Q2039846): Difference between revisions

Let \(K\) and \(L\) be two centrally symmetric, convex bodies in \(\mathbb R^{n}\). By \(K^{\circ}\) we denote the polar body of \(K\). The covering number \(N(K,L)\) is the least number so that \(K\) can be covered by that number of translates of \(L\). \textit{V. D. Milman} showed [C. R. Acad. Sci., Paris, Sér. I 302, 25--28 (1986; Zbl 0604.52003)] that there is a linear map \(T:\mathbb R^{n}\to\mathbb R^{n}\) so that \[ N(T(K),B_{2}^{n}),\ N((T(K))^{\circ},B_{2}^{n}),\ N(B_{2}^{n},T(K)),\ N(B_{2}^{n},N((T(K))^{\circ})\leq e^{c\cdot n}. \] This had been refined by \textit{G. Pisier} [J. Reine Angew. Math. 393, 115--131 (1989; Zbl 0692.46006)]: for any convex body \(K\) in \(\mathbb R^{n}\) and for any \(\alpha>\frac{1}{2}\), there is a linear map \(T_{\alpha}\) such that, for all \(t\geq1\), \[ N(T_{\alpha}(K),t B_{2}^{n}),\ N((T_{\alpha}(K))^{\circ},t B_{2}^{n}) \leq\exp\left(\frac{P_{\alpha}^{\frac{1}{\alpha}}n}{t^{\frac{1}{\alpha}}}\right), \] where \[ P_{\alpha}\leq\frac{c}{\sqrt{\alpha-\frac{1}{2}}} \] for \(\alpha\to\frac{1}{2}\). The authors refine these estimates further: \(G_{n,n-k+1}\) denotes the Grassmann manifold of \(n-k+1\)-dimensional subspaces of \(\mathbb R^{n}\) and \(\sigma_{n,n-k+1}\) its Haar measure. For every centrally symmetric, convex body \(K\) and for every \(\alpha>\frac{1}{2}\), there is a linear map \(T_{\alpha}\) such that, for all \(k=1,\dots,n\), \[ \sigma_{n,n-k+1}\left\{F\in G_{n,n-k+1} : \operatorname{diam}(T_{\alpha}(K)\cap F)>P_{\alpha} \left(\frac{n}{k}\right)^{\alpha} \right\} \leq e^{-c\cdot k}. \] The same inequalities also hold for \((T_{\alpha}(K))^{\circ}\). In another theorem, the authors prove the same type of inequalities for projections instead of subspaces.