Remarks on Minkowski symmetrizations (Q2702145)

It is known from previous work of \textit{J. Bourgain} et al. [Lect. Notes Math. 1317, 44-66 (1988; Zbl 0645.52001)] that with high probability a given (symmetric) convex body \(K\) in \({\mathbb R}^n\) can be transformed into a convex body \(\hat{K}\) by \(c n\log n\) successive random Minkowski symmetrizations with respect to independently and uniformly chosen random directions (hyperplanes through the origin) such that \(\hat{K}\) is close to a ball up to a factor. Here \(c\) is a universal constant. Let \(a(K)\) denote the circumradius of \(K\), and let \(M^*(K)\) be half of the mean width of \(K\). Using results and methods of the paper by Bourgain et al., the author refines the result stated above by showing that already \(cn\log(a(K)/M^*(K))\) random Minkowski symmetrizations suffice and in fact are needed, in a sharp sense. It is also proved that merely \(cn\log\log n\) suitably chosen (deterministic) directions are required to achieve that \(\hat{K}\) is close to a ball with the same precision of approximation. The idea of the proof of the latter fact is to use deterministic choices as well as the existence of suitable random choices. For the entire collection see [Zbl 0949.00025].