Small ball estimates for quasi-norms (Q501833)

A \textit{small ball probability} bound on \({\mathbb R}^n\) concerns the probability that a random vector \(X\) falls within a small ball defined in terms of some quasi-norm. In [the first and third author, Math. Ann. 350, No. 4, 953--972 (2011; Zbl 1229.46009)], it was discovered that if \(\| \cdot \|\) was a quasi-norm on \({\mathbb R}^n\), with unit ball \(K\), and if \(X\) had characteristic function \(\phi_X\), then if \(t > 0\), \[ {\mathbb P}(\| X \| \leq t) \leq |K| \left(\frac t{2\pi}\right)^n \int_{{\mathbb R}^n} |\phi_X(\xi)| \, d\xi. \] This result is tightened in order to establish an inequality involving a more specific random vector. Suppose that \(X\) is a linear combination of some fixed vectors -- but that the coefficients are i.i.d. random variables -- and impose some restrictions on the vectors, the i.i.d. random variables, and on \(t\). This article states an inequality of the form \[ {\mathbb P}(\| X \| \leq t) \leq \frac{|K|}{\gamma_n(K)} C^n \left( (c_1t)^n + c_2 \right), \] where \(\gamma_n(K)\) is the Gaussian measure of \(K\), and \(C\), \(c_1\), and \(c_2\) are combinations of parameters described in the article, thus tightening a variant of a result of [\textit{M. Rudelson} and \textit{R. Vershynin}, Commun. Pure Appl. Math. 62, No. 12, 1707--1739 (2009; Zbl 1183.15031)]. This article is largely computational, largely but not entirely self-contained, and therefore relatively accessible.