On stability of the B. Gnedenko characterization (Q1286620)

Let \(X_1,X_2\) be independent nondegenerate random variables. The linear statistics \(L_1=a_1X_1+ a_2X_2\) and \(L_2=b_1X_1+ b_2X_2\) are independent if and only if \(X_1,X_2\) are normally distributed. The above characterization, known as the Gnedenko characterization, is examined with respect to its stability, especially when the independence assumption is relaxed. Namely assuming that \(L_1,L_2\) to be `almost independent' under the condition \(a_1b_2 a_2b_2<0\), the random variables are almost normal. The author introduces the notion of `nonflow of a considerable mass to infinity' which provides simpler conditions for examining the stability of a characterization than those proposed by Zolotarev and others in this topic. Special cases of the above stability considerations are examined. For the case when \(a_1=a_2= b_1=1\) and \(b_2=-1\) (known as Bernstein's characterization) specific results are given.