Integral Kašin splittings (Q1425645)

A Kashin splitting with constant \(C\) in \(\mathbb{R}^{2n}\) is an orthogonal decomposition \(\mathbb{R}^n = E \oplus E^{\perp}\) where \(\dim(E) = n\) and with the following property: for any \(x \in E \cup E^{\perp}\), \[ \| x \| _1 \leq \| x \| _2 \leq C \| x \| _1 \] where for \(x \in \mathbb{R}^{2n}\) we put \(\| x \| _p = \left( \frac{1}{2n} \sum | x_i| ^p \right)^{\frac{1}{p}}\). The existence of a Kashin splitting for any dimension \(n\) with \(C > 0\) being a universal constant is a classical result due to \textit{B. S. Kashin} [Math. USSR, Izv. 11, 317-333 (1977; Zbl 0378.46027)]. Since then, several existence proofs were found, all of them are probabilistic in nature. No explicit construction of a Kashin splitting is currently known. This main result of the present paper is that for every \(n\), there exists a Kashin splitting of \(\mathbb{R}^{2n}\) where the subspace \(E \subset \mathbb{R}^{2n}\) is spanned by \(n\) vectors with integral entries. Furthermore, the absolute values of these integers do not exceed \(e^{ck}\), for some universal constant \(c > 0\). A central ingredient in the proof is a formula for \( \sup_{0 \neq x \in E} \frac{\| x \| _2}{\| x \| _1}\) in terms of determinants of square submatrices of the \(n \times 2n\) matrix whose rows span \(E\). Using this formula, the author analyses subspaces spanned by the rows of an \(n \times 2n\) matrix whose entries are independent standard Gaussians, and reproves the known result that such subspaces (which actually distribute uniformly over the Grassmanian) form a Kashin splitting with high probability. The author concludes his result with approximating Gaussian matrices by matrices with integral entries. A further investigation in this direction was carried out in \textit{G. Schechtman} [Isr. J. Math. 139, 337--347 (2004; Zbl 1066.46010)]. Schechtman considered a random \(n \times 2n\) matrix of the form \([\sqrt{n} I, A ]\) where \(I\) is the \(n \times n\) identity matrix, and \(A\) is an \(n \times n\) matrix whose entries are independent \(\pm 1\)-valued symmetric random variables. He proved that with high probability, the rows of such a matrix span a subspace \(E\) which is a Kashin splitting.