Dvoretzky type theorems for multivariate polynomials and sections of convex bodies (Q540007): Difference between revisions

The main result of this interesting paper is the following ``Dvoretzky-type'' result: For any \(d, k \in \mathbb N\) with \(d\) even, there is \(n(d,k) \in \mathbb N\) such that for every \(n \geq n(d,k)\) and for every homogeneous polynomial \(f:\mathbb R^n \to \mathbb R\) of degree \(d,\) there is a subspace \(V \subset \mathbb R^n, \;\dim\;V = k,\) such that \(f|_V\) is proportional to \((x_1^2 + ... + x_n^2)^{d/2}.\) This result is known as the Gromov-Milman conjecture [\textit{V. D. Milman}, Lect. Notes Math. 1317, 283--289 (1988; Zbl 0657.10020)]. (The original conjecture also stated that \(n(d,k)\) is of the order \(k^d;\) the authors note that their proof, which uses the Borsuk-Ulam theorem, does not settle this.) When \(d\) is odd the result can be found in [\textit{B. J. Birch}, Mathematika, Lond. 4, 102--105 (1957; Zbl 0081.04501)], although there is more recent work as well as estimates on \(n(d,k)\) by \textit{P. Hájek} and the reviewer in [Arch. Math. 86, No. 6, 561--568 (2006; Zbl 1106.46029)]. In fact, the estimates provided in the previous reference are improved here. Among other things, the authors show (Theorem 5) that for any \(d, k, m \in \mathbb N\) with \(d\) odd, there is \(n(d,k,m) \in \mathbb N\) such that for any \(n \geq n(d,k,m)\) and any odd polynomial map \(f:\mathbb R^n \to \mathbb R^m,\) all of whose coordinate functions have degree at most \(d,\) there is a \(k\)-dimensional subspace of \(\mathbb R^n\) that \(f\) maps to \(0.\) An explicit value of \(n(d,k,m)\) is given, as is a complex version of this result (Theorem 7) with a better bound. The authors discuss a stronger conjecture involving the Grassmannian of subspaces of \(\mathbb R^n\) of dimension \(k\). Although this conjecture is known to be false in general, the authors show that it does hold in certain cases such as \(k = 2p^\alpha\) where \(p\) is a prime number. As a consequence, they prove that for such \(k\) and all \(m,\) the following holds for sufficiently large \(n\): For any \(m\) convex bodies \(K_1, \dots, K_m\) in \(\mathbb R^n,\) there is a \(k\)-dimensional subspace \(L \subset \mathbb R^n\) such that the orthogonal projections of each \(K_i\) onto \(L\) has a Euclidean ball as its John ellipsoid.