Estimates of the proximity of a distribution to the normal law (Q920471): Difference between revisions

Let X be a r.v. with mean 0, variance 1 and E \(| X|^ 3=\beta <\infty\). Denote \(F(x)=P(X<x)\) and let \(\Phi\) (x) be the standard normal distribution function on \({\mathbb{R}}^ 1\). The paper is devoted to the problem of uniform and non-uniform estimation of the difference \(D(x):=F(x)-\Phi (x)\). The main results are: The smallest absolute constant c such that \[ \sup_{F}\sup_{x\in {\mathbb{R}}^ 1}| D(x)| \leq c\beta \text{ is equal to } c=0.370352... \] (X symmetric r.v. then \(c=0.341345...)\) and the smallest absolute constant u such that \[ \sup_{F}\sup_{x\in {\mathbb{R}}^ 1}| x|^ 3| D(x)| \leq u\beta \text{ is equal to } u=1 \] (X symmetric r.v. then \(u=0.5)\). In the uniform error estimation the equality is achieved iff X is a Bernoulli r.v., however in the non- uniform error estimation the equality is not achieved. Moreover in the symmetric case exact bounds for F(x) with fixed x and \(\beta\) are obtained. In the general case exact bounds for F(x) are obtained if \(-1\leq x\leq -(1+\sqrt{2})^{-1/2}\) or \(x=0\).