No label defined (Q4944243)

Let \(\Phi: =\{\varphi: (0,\infty) \to(0,\infty): \varphi(t)/t \uparrow \infty\) as \(t\uparrow \infty\}\), and \(\Phi_s: =\{\varphi \in\Phi: \sum^\infty_{n=1} 1/\varphi(n) <\infty\}\), and put \(\varphi (0)=0\) for \(\varphi \in\Phi\). A sequence \((X_n:n= 0,1,\dots)\) of integrable random variables is called Cesàro uniformly integrable (resp. Cesàro strongly uniformly integrable) if there exists \(\varphi\) in \(\Phi\) (resp. \(\varphi\) in \(\Phi_s)\), with \(\sup_{n\geq 1}{1\over n}\sum^n_{i=1} E(\varphi (|X_i|)) <\infty\). The authors obtain the following results: Let \((X_n)\) be a sequence of non-negative integrable random variables with \(E(X_iX_j)\leq E(X_i) E(X_j)\) for \(i\neq j\). (i) If \((X_n)\) is Cesàro uniformly integrable, then \({1\over n} \sum^n_{i=1} (X_i-E(X_i)) \to 0\) in \(L_1\) and hence in \(P\)-measure. (ii) If \((X_n)\) is Cesàro strongly uniformly integrable, then \({1\over n}\sum^n_{i=1} (X_i-E(X_i)) \to 0\) \(P\)-a.s. This theorem yields WLLN and SLLN for pairwise independent random variables if it is applied to the positive and the negative parts of the given random variables separately. Finally, the authors give an example of a sequence \((X_n)\) of non-negative uncorrelated random variables which is Cesàro strongly uniformly integrable (hence it obeys SLLN), but these random variables are not pairwise independent; morover, \(\text{Var}(X_n)\geq n^2\).