A sharp form of the Cramér-Wold theorem (Q2641421)

The well known Cramér-Wold theorem states that a Borel probability measure \(P\) on \(\mathbb{R}^d\) is uniquely determined by its one-dimensional projections. The authors consider the problem: How large a subset of these projections is really needed to determine \(P\)? For two probability measures \(P\) and \(Q\), let \({\mathcal E}(P,Q)\) be the set of all \(x \in \mathbb{R}^d\) for which projections of \(P\) and \(Q\) onto \(\langle x\rangle\), the one-dimensional subspace generated by \(x\), coincide. The main positive result of the paper: Let the absolute moments \(m_n:=\int \| x\| ^n\,dP(x)\) be finite, \(\sum_nm_n^{-1/n}=\infty\) and the set \(\mathcal{E}(P,Q)\) is not contained in the zero set of any homogeneous polynomial. Then \(P=Q\). Characteristic functions play the central role in the proof. Examples show that conditions of the theorem are necessary. The authors also prove similar theorems for Hilbert spaces.