Wiener's `closure of translates' problem and Piatetski-Shapiro's uniqueness phenomenon (Q640793)

For \(1 \leq p < \infty\), a sequence \({\mathbf c} = (c_n) \in \ell_p ({\mathbb Z})\) is said to be \textit{cyclic} in \(\ell_p ({\mathbb Z})\) if the linear span of the translates of \({\mathbf c}\) is dense in \(\ell_p ({\mathbb Z})\). \textit{N. Wiener} [``Tauberian theorems'', Ann. Math. (2) 33, 1--100 (1932; Zbl 0004.05905)] proved, as a tool towards his Tauberian theorems, that: (1) a sequence \({\mathbf c} = (c_n)\) is cyclic in \(\ell_2 ({\mathbb Z})\) if and only if its Fourier transform \(\widehat {\mathbf c} (t) = \sum_{n \in {\mathbb Z}} c_n \text{e}^{int}\) is nonzero almost everywhere; (2) \({\mathbf c}\) is cyclic in \(\ell_1 ({\mathbb Z})\) if and only if \(\widehat {\mathbf c}\) is nonzero everywhere (though he actually stated these results for \(L^2 ({\mathbb R})\) and \(L^1 ({\mathbb R})\)), and he conjectured that the cyclicity of vectors \({\mathbf c}\) in \(\ell_p ({\mathbb Z})\) could be characterized by the smallness of the set \(Z_{\widehat {\mathbf c}} = \{ t \in {\mathbb T}\,;\;\widehat {\mathbf c} (t) = 0\}\) of the zeros of the Fourier transform \(\widehat {\mathbf c}\) of \({\mathbf c}\), at least for \(1 \leq p \leq 2\) (in their book, [\textit{J.-P. Kahane} and \textit{R. Salem}, Ensembles parfaits et séries trigonométriques, Actualités Scientifiques et Industrielles 1301, Paris, Hermann \& Cie (1963; Zbl 0112.29304)] referred to this question as posed by A. Beurling in 1947). In the paper under review, the authors disprove this conjecture: they show that, for every \(1 < p < 2\), there exist \({\mathbf c}_1, {\mathbf c}_2 \in \ell_1 ({\mathbb Z})\) such that \(Z_{\widehat {\mathbf c}_1} = Z_{\widehat {\mathbf c}_2}\), but \({\mathbf c}_1\) is cyclic in \(\ell_p ({\mathbb Z})\), even though \({\mathbf c}_2\) is not. A slightly weaker version was announced by the authors previously [``No characterization of generators in \(\ell ^p\) \((1 < p < 2)\) by zero set of Fourier transform'', C. R. Math. Acad. Sci. Paris 346, No. 11--12, 645--648 (2008; Zbl 1155.46015)]. Actually, this result is a straightforward consequence of the following theorem (Theorem 1): for every \(1 < p < 2\), there exists a compact set \(K \subseteq {\mathbb T}\) such that: (i) if \(\sum_{n \in {\mathbb Z}} |n|^\varepsilon |c_n| < \infty\) for some \(\varepsilon > 0\) and \(\widehat {\mathbf c} = 0\) on \(K\), then \({\mathbf c}\) is not cyclic in \(\ell_p ({\mathbb Z})\); (ii) there exists \({\mathbf c}_1 \in \ell_1 ({\mathbb Z})\) such that \(\widehat {\mathbf c}_1 = 0\) on \(K\) and \({\mathbf c}_1\) is cyclic in \(\ell_p ({\mathbb Z})\). To prove this theorem, the authors look at the ``other side'' of the Fourier transform. Recall that every bounded sequence \({\mathbf c} \in \ell_\infty ({\mathbb Z})\) defines a distribution \(S\) on \({\mathbb T}\) such that \(\widehat S = {\mathbf c}\), called a \textit{pseudo-measure}. This pseudo-measure is called a \textit{pseudo-function} if \(\widehat S \in c_0 ({\mathbb Z})\). For \(1 \leq q < \infty\), let \(A_q ({\mathbb T})\) be the space of pseudo-functions \(S\) such that \(\widehat S \in \ell_q ({\mathbb Z})\). As a consequence of their Theorem 1, they get that for every \(q > 2\), there exists a compact set \(K \subseteq {\mathbb T}\) such that: (i') there exists a nonzero \(S \in A_q ({\mathbb T})\) such that \(\text{supp}\,S \subseteq K\); (ii') there is no nonzero measure \(\mu \in A_q ({\mathbb T})\) such that \(\text{supp}\,\mu \subseteq K\) (this result was announced by the authors in [``Piatetski-Shapiro phenomenon in the uniqueness problem'', C. R. Math. Acad. Sci. Paris 340, No. 11, 793--798 (2005; Zbl 1063.43006)]). Actually, (i') is equivalent to (i). This corollary gives a link to the classical notion of sets of uniqueness and sets of multiplicity. Indeed, it is known that a compact set \(K \subseteq {\mathbb T}\) is a \textit{set of uniqueness} if and only if no nonzero pseudo-function is supported by \(K\). Hence the corollary says that a counterexample ``must'' be a set of multiplicity of a particular type which were constructed for the first time by I. Piatetski-Shapiro in 1954: compact sets of multiplicity which support no nonzero Rajchman measure \(\mu\) (i.e., \(\widehat \mu \in c_0 ({\mathbb Z})\)). Later, in 1973, T. Körner and R. Kaufman, independently, constructed such sets which are moreover \textit{Helson sets}. It appears that such sets are the good ones: the authors show that for every (compact) Helson set \(K\), there is a function \(g \in A ({\mathbb T})\) (the Wiener algbra) vanishing on \(K\) and cyclic in \(A_p ({\mathbb T})\) for every \(p > 1\). Hence to get Theorem 1, it suffices to show (Theorem 3): For every \(q > 2\), there exists a (compact) Helson set \(K \subseteq {\mathbb T}\) which supports a nonzero pseudo-function \(S \in A_q ({\mathbb T})\). This is the main result of the paper and its proof requires several lemmas and tools: Rudin-Shapiro polynomials, a quantitative version of a lemma of Kahane, a generalization of the Azuma inequality (which is itself a generalization of the Bernstein inequality for the sums of independent random variables), measures \(\mu_s\), \(0 < s < 1\), constructed as Riesz-type products; construction of ``almost independent'' random variables, exponential estimates of deviation for the empirical mean of these variables with respect to \(\mu_s\) and exponential \(L^2\) estimate. At the end of the paper, the authors also explain how their Theorem 1 gives the corresponding version for \(L_p ({\mathbb R})\), make some remarks and ask some questions.