On the uncertainty principle for M. Riesz potentials (Q1426907)

Let \(\mu\) be a Borel charge (real measure) in \(\mathbb R^d\) satisfying \[ \int_{\mathbb R^d} {{d| \mu| (t)}\over{1+| t| ^{d-\alpha}}} < +\infty \tag{(1)} \] for an \(\alpha \in (0,d)\). We denote by \(U^{\mu}_{\alpha}\) the M.\ Riesz potential of \(\mu\) of order \(\alpha\), \[ U^{\mu}_{\alpha}(s)= \int_{\mathbb R^d} {{d \mu (t)}\over{| t-s| ^{d-\alpha}}}, \qquad s \in \mathbb R^d. \] (Note that \(U^{\mu}_{\alpha}(s) < +\infty\) \(m\) -- a.e. in \(\mathbb R^d\), where \(m=m_d\) is Lebesgue measure in \(\mathbb R^d\).) If \(\mu\) is \(m\)-absolutely continuous, i.e., \(\mu=fm\), then we write \(U^{\mu}_{\alpha}=U^{f}_{\alpha}\). The following result is due to \textit{M.\ Riesz} [Acta Litt. Sci. Szeged 9, 1--42 (1938; Zbl 0018.40704 )]: If \(\alpha\) is not an even integer, then \(| \mu| \) and \(U^{\mu}_{\alpha}\) cannot vanish on the same nonvoid open set unless \(\mu=0\). For \(d=1\) this ``uncertainty principle'' applies to sets \(E \subset \mathbb R\) of positive length (not necessarily open) provided \(\mu=f m\) and \(f\) is \textit{sufficiently smooth}. To give a precise statement we need the following definition: a charge \(\mu\) is called \textit{small of order \(\gamma >0\) at the point \(s_0 \in R\)} if \[ \int_{(s_0-1,s_0+1)} {{d| \mu| (t)}\over{| t-s_0| ^{\gamma}}} < +\infty. \] If \(\gamma\geq1\) and \(\mu\) is small of order \(\gamma\) at any point of a Borel set \(E\), then \(| \mu| (E)=0\) (see p. 513 of [\textit{V.\ Havin} and \textit{B.\ Jöricke}, ``The uncertainty principle in harmonic analysis'' (1994; Zbl 0827.42001)] for a proof), but not viceversa: \(| \mu| (E)=0\) does not imply the \(\gamma\)--smallness of \(\mu\) on \(E\). Theorem 1. (The uncertainty principle for the Riesz potentials on the line)\ Suppose \(E \subset \mathbb R\) is a Borel set, \(m(E) >0\), \(\alpha \in (0,1)\), and \(\mu\) satisfies (1) with \(d=1\). If \(\mu\) is small of order \(2-\alpha\) at any point of \(E\) and \(U^{\mu}_{\alpha}| _{E} =0\), then \(\mu=0\). (Here \(m=m_1\).) This theorem was proved by \textit{V.\ P. Khavin} [Sov. Math. Dokl. 25, 694--698 (1982); translation from Dokl. Akad. Nauk SSSR 264, 559--563 (1982; Zbl 0509.31009)]. Corollary. Let \(\mu\), \(\alpha\) and \(E\) be as in Theorem 1. Suppose \(\mu=f m\) in a neighborhood of \(E\), \(f\) being a Hölder function of order \(1-\alpha+\varepsilon\), \(\varepsilon >0\). If \(f| _E= U^{\mu}_{\alpha}| _E=0\) and \(m(E) >0\), then \(\mu = 0\). The smallness of \(\mu\) of order \(2 -\alpha\) (instead of just \(| \mu| (E)=0\)) in Theorem 1 and the Hölder condition in the Corollary may look strange and give the impression of being redundant technicalities due to the (possibly inadequate) method of proof. However, in the paper under review the authors give a \textit{negative answer} to the following question: Is it possible to drop the Hölder condition in the Corollary, replacing it by the mere continuity of \(f\)? Their main result is the following theorem. Theorem 2. For any \(\alpha \in (0,1)\) there exists a nonzero function \(f \in C(\mathbb R)\) such that \[ m(\{x:f(x)=0\}\cap\{x:U_{\alpha}^{f}(x)=0\})>0. \] Thus the smallness condition imposed on \(\mu\) in Theorem 1 and the Hölder condition imposed on \(f\) on its Corollary are essential. A multivariate version of Theorem 2 is sketched at the end of the article.