Bounded orthogonal systems and the \(\Lambda\) (p)-set problem (Q1121477): Difference between revisions

Assume that \(\phi_ 1,...,\phi_ n\) is a sequence of mutually orthogonal functions, for example a sequence of unitary characters in the dual group \(\Gamma\) of a compact Abelian group G. As in the example, assume further that all functions \(\phi_ j\) are uniformly bounded by 1, and let \(2<p<\infty\). The author proves the following amazing theorem: There always exists a subset \(S\subset \{1,...,n\}\) of cardinality \(| S| >n^{2/p}\), such that \[ \| \sum_{i\in S}a_ i\phi_ i\|_{L^ p}\leq C(p)(\sum_{i\in S}| a_ i|^ 2)^{1/2} \] for all scalar sequences \(\{a_ i\}_ i\). Here C(p) is a constant which only depends on p. In fact he shows that, in a probabilistic sense, the above estimate holds for a generic randomly chosen set S. Let me recall that, if G and \(\Gamma\) are as above and \(p>2\), a subset \(\Lambda\) \(\subset \Gamma\) is a \(\Lambda\) (p)-set, if \(L^ p_{\Lambda}(G)=L^ 1_{\Gamma}(G)\), or, equivalently, \(L^ p_{\Lambda}(G)=L^ 2_{\Lambda}(G)\), where \(L^ p_{\Lambda}(G)\) is the closure in \(L^ p(G)\) of the linear space spanned by the characters in \(\Lambda\). As a consequence of the main theorem, the author can show that for every \(2<p<\infty\) there exists a \(\Lambda\) (p)-subset of \(\Gamma ={\mathbb{Z}}\) which is not a \(\Lambda\) (r)-set for any \(r>p\), which solves the so-called \(\Lambda\) (p)-set problem. So far, by a result of Rudin, this had only been known for even integers \(p=2m\geq 4.\) The ingenious proof of the main theorem is based on a subtle probabilistic estimate and a related entropy estimate. The exposition is self-contained.