Geometry of families of random projections of symmetric convex bodies. (Q5957692)

This paper provides a profound study of the average distance between the orthogonal projections of two symmetric convex bodies onto two independent uniform random subspaces. Here we can only describe the following very special case. Let \(d(K,L)\) denote the Banach-Mazur distance of two centrally symmetric convex bodies \(K,L\subset{\mathbb R}^N\), i.e., the smallest constant \(c>0\) such that \(K\subset TL\subset c K\) for some linear transformation \(T\) of \({\mathbb R}^N\). Further, assume that the Euclidean unit ball \(B^N_2\) in \({\mathbb R}^N\) is the ellipsoid of minimal volume containing \(K\). Fix \(\lambda\in (0,1)\), \(1\leq n\leq \lambda N\), \(\epsilon\in(0,1/2)\) and set \(m:=[(1/2-\epsilon)n ]\). Then there is a constant \(c=c(\lambda,\epsilon)\) such that for the averages one finds the inequality \[ {\mathbb E}_{E,F}d(P_EK,P_{F}K)\geq c\left({\mathbb E}_Ud(P_UK,B^m_2)\right)^2. \] The mean value on the left is taken with respect to orthogonal projections \(P_E,P_F\) onto independent uniform random \(n\)-dimensional subspaces \(E,F\) of \({\mathbb R}^N\) and on the right with respect to orthogonal projections \(P_U\) onto uniform random \(m\)-dimensional subspaces \(U\) of \({\mathbb R}^N\). Note that this estimate is the deep converse estimate of an obvious inequality. Examples show that the drop of the dimension \(m\) to a proportion of \(n\) less than \(1/2\) cannot be avoided. The results established by the authors are more general in several respects. In particular, they are obtained by estimating \({\mathbb E}_{E,F}d(P_EK,P_{F}K)\) from below by the averages of some volumetric invariants which generalize volume ratios. In addition, the latter averages are estimated from below by \({\mathbb E}_Ud(P_UK,B^m_2)\). An important step in the proof and results of independent interest are corresponding estimates for Gaussian projections.