An exponential inequality for negatively associated random variables (Q5902217): Difference between revisions

Let \(\{X_n, n\geq 1\}\) be sequence of identically distributed negatively associated random variable with \(Ee^{\delta(X_1)}<\infty\) for some \(\delta >0\). The main statements are as follows. If \(\epsilon_n=\sqrt{2\delta E|X_1|^2C_n/n}\) where \(0<c_n\leq\Big(\frac{eE|X_1|^2n}{8\delta}\Big)^{1/3}\), then \[ P\Bigg( \frac{1}{n}\Bigg|\sum\limits^n_{i=1} (X_i-EX_i)\Bigg | >3\epsilon_n\Bigg) \leq 2\Bigg(1+\frac{Ee^{\delta|X|}}{\delta^3eE|E|X_1|^2c_n}\Bigg)e^{-\delta c_n}. \] If \(\epsilon_n=\sqrt{2\delta eE|X_1|^2\log n/n}\), then \[ \sum\limits_{n=1}^\infty P\Bigg(\frac{1}{n}\Bigg |\sum\limits_{i=1}^n(X_i-EX_i)\Bigg| >3\epsilon_n\Bigg)<\infty . \] The Borel-Cantelli lemma implies that \(\frac{1}{n}\sum^n_{i=1} (X_i-EX_i)<\infty\) almost surely with rate \((3\epsilon_n)^{-1}=0(1) n^{1/2}(\log n)^{-1/2}\). References on similar results for negatively and for positively associated random vectors are given in the Introduction. Section~2 contains preliminary lemmas. The main results are stated and proved in Section~3. The volume of the paper is 7 pages. The list of references contains 15 positions.