On the stability of a characterization by identically distributed statistics (Q1781671)

The authors prove the following result. Assume that given \(\varepsilon>0\) and \(\alpha\in (0,2]\), \[ \sup_t | f(t)-f^{n}(n^{-1/\alpha}t)| \leq \varepsilon, \quad n=2,3, \] where \(f(t)\) is the characteristic function of the random variable. If \(\alpha \neq 1\), then there exists a stable characteristic function \(g(t)\) and a constant \(C_1>0\) such that \(\sup_t| f(t)-g(t)| \leq C_1\varepsilon^{1/5}\). If \(\alpha=1\) and \(f(t_0)\in \mathbb{R}\), for some non-zero \(t_0\), then there exist constants \(A\in\mathbb{R}\) and \(C_2>0\) such that \[ \sup_t| f(t)-\exp(-| A|\, | t| )| \leq C_2\varepsilon^{1/4}. \] In case \(\varepsilon=0\) the result reduces to a known theorem due to Paul Lévy.