Tail bounds for sums of independent two-sided exponential random variables (Q6150872)

Let \(X_1,X_2,\ldots\) be independent and identically distributed Laplace random variables, each with density function \(\frac{1}{2}e^{-|x|}\) for \(x\in\mathbb{R}\), and let \(S=\sum_{i=1}^na_iX_i\) for some positive weights \(a_i\). The main result of the present paper is the two-sided concentration inequality \[ \frac{1}{57}\frac{1}{\sqrt{\alpha t}}\exp\left(-\alpha t\right)\leq\mathbb{P}\left(S>t\sqrt{\text{Var}(S)}\right)\leq\exp\left(-\frac{\alpha^2}{2}h\left(\frac{2t}{\alpha}\right)\right) \] for \(t>1\), where \(\alpha=\frac{\sqrt{2\sum_{i=1}^na_i^2}}{\max_ia_i}\) and \(h(u)=\sqrt{1+u^2}-1-\log\frac{1+\sqrt{1+u^2}}{2}\). As \(t\to\infty\), both the upper and lower bounds are of order \(e^{-\alpha t+o(t)}\). The proof is based on a representation as a Gaussian mixture. The authors also give an analogous result without the assumption that the \(X_i\) have a Laplace distribution, and use their concentration inequality to give estimates for moments of \(S\). For the entire collection see [Zbl 07730209].