Superconcentration, and randomized Dvoretzky's theorem for spaces with 1-unconditional bases (Q1675908)

The author shows that, if a centrally-symmetric convex body \(K\) is 1-unconditional and is in the \(\ell\)-position, then a random \((c\varepsilon \ln n /\ln(1/\varepsilon))\)-dimensional section of \(K\) is \((1+\varepsilon)\)-Euclidean. In the \(1\)-unconditional case, this significantly improves the dimension of a \((1+\varepsilon)\)-Euclidean section in Milman's proof of the Dvoretzky theorem, where the dimension \((c\varepsilon^2 \ln n /\ln(1/\varepsilon))\) was first obtained by \textit{V. D. Mil'man} [Funct. Anal. Appl. 5, 288--295 (1972; Zbl 0239.46018); translation from Funkts. Anal. Prilozh. 5, No. 4, 28--37 (1971)] and later the logarithmic factor in \(\varepsilon\) was removed by \textit{Y. Gordon} [Isr. J. Math. 50, No. 4, 265--289 (1985; Zbl 0663.60034)]. Remarkably, the author uses the \(\ell\)-position, while in Milman's proof the John position was used. The result also improves \textit{G. Schechtman}'s estimate \((c\varepsilon \ln n /\ln^2(1/\varepsilon))\) [Adv. Math. 200, No. 1, 125--135 (2006; Zbl 1108.46011)] by showing that a random subspace satisfies the bound. The author proves the sharpness of his result about a random Euclidean section of a 1-unconditional body in \(\ell\)-position as well as the sharpness of the bound \(c\varepsilon^2 \ln n\) for a body in the John position. One of the main new ingredients is related to the concept of superconcentration, recently introduced by \textit{S. Chatterjee} [Superconcentration and related topics. Cham: Springer (2014; Zbl 1288.60001)]. The author proves that the norm corresponding to a 1-unconditional body in \(\ell\)-position raised to a power \(p\leq \ln n\) is \((C/\ln n)\)-superconcentrated in the Gauss space. This implies the main result.