Abstract:
We consider the sequence (1) of compositions of distribution functions satisfying the condition (2). Let truncated variances of components be bounded from below by a positive constant. It is proved that the well-known estimate
$$
\max_{i=1,2}\inf_{G\in N}\sup_x|F_n^{(i)}(x)-G(x)|=O\biggl(\frac1{\sqrt{-\ln\varepsilon_n}}\biggr)
$$
(where $N$ is the set of all normal distribution functions) is unimprovable.