Sidonicity and variants of Kaczmarz's problem (Q1687896)

It is well-known, as shown by Rudin, that every Sidon set \(\Lambda\) in a discrete Abelian group is a \(\Lambda (p)\)-set for all \(p < \infty\) and moreover, for \(2 \leq p < \infty\), we have \(\| f \|_p \leq C \, {\sqrt p} \, \|f \|_2\) for every trigonometric polynomial \(f\) with spectrum in \(\Lambda\). \textit{G. Pisier} [ Ec. Polytech., Cent. Math., 1977-Exposes No. 12, 13, 33 p. (1978; Zbl 0388.43009)] showed spectacularly, using Gaussian processes (and Rider's theorem), that the converse holds. Other proofs are given by the first author [Ann. Inst. Fourier 35, No. 1, 137--148 (1985; Zbl 0578.43008)] and \textit{G. Pisier} [Proc. Semin., Torino and Milano 1982, Vol. II, 911--944 (1983; Zbl 0539.43004)], for example. In the paper under review, the authors study this problem when the characters of the compact Abelian group are replaced by any uniformly bounded orthonormal system (OS in short). It appears that this problem was raised by S. Kaczmarz in the Scottish book, as Problem 130. If \((\phi_j)\) is an OS of complex-valued functions on a probability space \(\Omega\), this system is said to be Sidon if, for all complex numbers: \(\big\| \sum_j a_j \phi_j (\omega) \big\|_\infty \geq c \sum_j |a_j|\), and it is said to satisfy the \(\psi_2\)-condition if \(\big\| \sum_j a_j \phi_j \big\|_p \leq C \, \sqrt{p} \big( \sum_j |a_j|^2\big)^{1/2}\) for all \(p \geq 2\), condition that is equivalent to \(\big\| \sum_j a_j \phi_j \big\|_{\psi_2} \leq C \big( \sum_j |a_j|^2\big)^{1/2}\), where \(\| \, . \, \|_{\psi_2}\) is the norm of the Orlicz space associated to the function \(\psi_2 (x) = \text{e}^{x^2} - 1\). First of all, the authors show that this \(\psi_2\)-condition does not imply Sidonicity (Theorem 1.2). However, they prove (Theorem 1.3) that this \(\psi_2\)-condition for a uniformly bounded OS \((\phi_j)\) implies that it is a \(5\)-tensor Sidon system, i.e. the OS \((\Phi_j)\) is Sidon, with \(\Phi_j = \phi_j \otimes \phi_j \otimes \phi_j \otimes \phi_j \otimes \phi_j\). A Rademacher-Sidon system (or Rider system) is a uniformly bounded OS \((\phi_j)\) such that \({\mathbb E}_\omega \big\| \sum_j r_j (\omega) a_j \phi_j \big\|_\infty \geq c \sum_j |a_j|\), where \((r_j)\) is a Rademacher sequence. Clearly, every Sidon OS is Rademacher-Sidon. \textit{D. Rider} [Duke Math. J. 42, 759--764 (1975; Zbl 0345.43008)] proved the converse when the OS is a set of characters. The authors prove (Theorem 1.4) that if the \(k\)-fold tensor product of an OS is Rademacher-Sidon, then this OS is itself Rademacher-Sidon and (Theorem 1.5) that every OS with the \(\psi_2\)-condition is Rademacher-Sidon. To prove Theorem 1.3 and Theorem 1.5, the OS \((\phi_j)\) is approximated by a martingale difference sequence and then Riesz product-type arguments are used. The authors give also a more sophisticated proof of Theorem 1.5, using the theory of stochastic processes (Preston's theorem), Talagrand's majorizing measure theorem [\textit{M. Talagrand}, Acta Math. 159, No.1--2, 99--149 (1987; Zbl 0712.60044)] and recent estimates on Bernoulli processes due to \textit{W. Bednorz} and \textit{R. Latała} [Ann. Math. (2) 180, No. 3, 1167--1203 (2014; Zbl 1309.60053)]. As a consequence of Theorem 1.5 and of the Elton-Pajor theorem, they get that every uniformly bounded OS with the \(\psi_2\)-condition contains a proportional subsystem which is Sidon. Note, as quoted at the end of the paper, that \textit{G. Pisier} [Math. Res. Lett. 24, No. 3, 893--932 (2017; Zbl 1385.43001)] showed that every uniformly bounded OS with the \(\psi_2\)-condition is actually a \(2\)-fold tensor Sidon system, and that being Rademacher-Sidon is equivalent to being \(4\)-fold Rademacher-Sidon.