Ergodic characterization of van der Corput sets (Q420970)

A set \(D\subset\mathbb N\) is called intersective if \(A-A\) contains an element of \(D\) for any set \(A\) of positive upper density, and is called a set of Poincaré recurrence if for any positive measure set \(A\) in a measure-preserving dynamical system \((T,\mu)\) there is some \(d\in D\) with \(\mu(A\cap T^{-d}A)>0\). \textit{A. Bertrand-Mathis} [Isr. J. Math. 55, 184--198 (1986; Zbl 0611.10032)] showed that a set is intersective if and only if it is a set of Poincaré recurrence. \textit{T. Kamae} and \textit{M. Mendès France} [Isr. J. Math. 31, 335--342 (1978; Zbl 0396.10040)] introduced the following related notion: a set \(D\) is called van der Corput whenever for any real sequence \((x_n)\), if all the sequences \((x_{n+d}-x_n)\) for \(d\in D\) are uniformly distributed modulo \(1\), then the original sequence \((x_n)\) is also uniformly distributed modulo \(1\). They showed that van der Corput sets are intersective, and \textit{J. Bourgain} [Isr. J. Math. 59, 150--166 (1987; Zbl 0643.10045)] later showed that the converse does not hold. Here it is shown that a set \(D\) is van der Corput if and only if for any Hilbert space \(H\), unitary operator \(U\), and \(x\in H\) with non-zero projection onto the kernel of \((U-I)\), there is some \(d\in D\) such that \((U^dx,x)\neq 0\).