Random sets of uniform convergence (Q1337836)

Denote by \(S_ N\) the family of all subsets \(\sigma\subset \{-N,- N+1,\dots, N\}\) and for some \(\sigma\in S_ N\) let \[ U(\sigma):= \sup\Bigl\| \sum_{n\in \sigma\cap[-s, s]} a_ n e^{2\pi inx}\Bigr\|_ \infty\Bigl/\Bigl\| \sum_{n\in \sigma} a_ n e^{2\pi inx}\Bigr\|_ \infty, \] where the supremum is taken with respect to all \(s\in N\) and all sequences \(\{a_ n: n\in \sigma\}\equiv 0\). As it is known, \(U(\sigma)\leq C\log N\) for all \(\sigma\) and \(N\) with some constant \(C\). Denote by \(S^ q_ N\) the subfamily of \(S_ N\), when the number of elements of \(\sigma\) equals \(q\), and by \(\nu\) the normalized counting measure on \(S^ q_ N\). The main result of the present paper reads as follows. Theorem 2. There exists an absolute constant \(b> 0\) such that for \(N= 2,3,\dots\) and \(1\leq q\leq N\), \[ \nu\left\{\sigma\in S^ q_{2N+1}: U(\sigma)\leq b \log\left(2+ {q\over \log N}\right)\right\}< {1\over N^ 2}. \] In particular, it follows that if for a random subset \(\sigma\in S^ q_ N\) we have \(q\gg\log N\), then \(U(\sigma)\to\infty\) as \(N\to\infty\) with high probability; and if \(q\geq N^ \varepsilon\) for some \(\varepsilon> 0\), then with high probability \(U(\sigma)\) has maximal growth order, that is, \(\log N\). Analogous results are proved for the Walsh system, as well.