Bemerkung über paarweise unabhängige zufällige Größen. (Remark on pairwise independent random variables) (Q1821423): Difference between revisions

Let \(\{\xi_ n\}^{\infty}_{n=1}\) be a sequence of random variables with \(E\xi_ n=0\). Consider the quantity \(I=\sum^{\infty}_{n=1}D^ 2(\xi_ n)\log^ 2n/n^ 2\), where \(D(\xi_ n)\) denotes the standard deviation of \(\xi_ n\). It is well known that, if I is finite and if \(\{\xi_ n\}\) is a sequence of orthogonal r.v.'s, then \((\xi_ 1+...+\xi_ n)/n\to 0\) a.s. (n\(\to \infty)\) [\textit{P. Revesz}, The laws of large numbers. (1968; Zbl 0203.504)]. The present author has shown that the condition that I be finite, is best in the following sense: if, for a sequence \(\{D_ n\}\) of real numbers such that \(D_ n/n\geq D_{n+1}/(n+1)\) (n\(\geq 1)\), I is finite, then on a suitable probability space we construct a sequence \(\{\xi_ n\}\) of orthogonal r.v.'s such that \(\overline{\lim}| \xi_ 1+...+\xi_ n| /n=\infty\) a.s. and \(D(\xi_ n)=D_ n\) for all n [the author, Period. Math. Hung. 2, 33-39 (1972; Zbl 0246.60017)]. In the present paper, he shows that even if we replace orthogonality by pairwise independence (less restrictive condition), the condition that I be finite is best in the above sense.