Error of normal approximation (Q794332)

For a sum of centered independent random vectors \(X_1, X_2,\ldots\); \(E| X_i|^3 \leq L\), \(E| X_i|^2 \leq 1\) taking values in the space \(\ell_{2k}\) \((k\) is natural) with a common covariance operator and \(Y\) a Gaussian centered random vector in \(\ell_{2k}\) with the same covariance operator the following result is proved: \[ P\{| a+S_n|<r\}-P\{| a+Y|<r\} \leq cLn^{-\frac{1}{2}}(1+| a|^{3(2k-1)}) \] for any \(a\in \ell_{2n}\), \(r\geq 0\), \(S_n=n^{-\frac{1}{2}}\sum^n_{i=1}X_i\), the constant \(c\) depends only on the structure of the covariance operator. The proof is based on the characteristic functions of the polynomials of the sum of random vectors. Interesting results are obtained in this direction, too.