Estimations of the decomposition stability into identical components (Q1907913)

Let \(F_1\), \(F_2\) and \(F\) be three probability distributions such that \(F= F_1* F_2\), where \(*\) means convolution. According to the famous Cramér's theorem on decomposition of a normal distribution, if \(F\) is normal then \(F_1\) and \(F_2\) are normal too. The authors investigate stability estimations in Cramér's theorem with the additional condition of the identity of components. They proved the following interesting result: Let \(F\) be a distribution function and \(\varepsilon= \rho(F* F,\Phi)<1\), where \(\rho\) means the uniform metric and \(\Phi\) is the standard normal distribution. If the median of \(F\) equals zero then there exists a constant \(C>0\) such that \(\rho(F, \Phi_{0,1/2})< C(\delta) \varepsilon^{1/4- \delta}\), where \(\delta\) is arbitrary positive and \(C(\delta)\) depends only on \(\delta\). Some generalizations of this result are also analyzed. As an illustration of the application of the estimations obtained a lower bound on the rate of convergence to stable laws is given.