A general law of precise asymptotics for products of sums under dependence (Q2266871)

Let \(\{X_{n},n\geq1\}\) be a strictly stationary, linear negative quadrant dependent sequence of random variables with \(EX_1=\mu>0\) and Var\(X_1=\sigma^2<\infty\), and set \(\sigma_1^2=1+2\sigma^{-2}\sum_{n\geq2}\text{Cov}(X_1,X_n)\) and \(S_n=X_1+\dots+X_n\). Let \(g(x)\) be a strictly increasing, positive and differentiable function on \([n_0,\infty[\) with \(\lim_{x\to\infty}g(x)=\infty\), and put \(\phi(x)=g'(x)/g(x)\), \(x\geq n_0\). Following somewhat standard procedures, under mild auxiliary conditions, the authors derive precise asymptotics such as \[ \sum_{n\geq n_0}\phi(n)P\left[\left(\frac{S_1\cdots S_n}{n!\mu^n}\right)^{\frac{\mu}{\sigma\sigma_1\sqrt{2n}}}\geq e^{\varepsilon g^{s}(n)}\right]\sim(-\log\varepsilon)/2s \] as \(\varepsilon\searrow 0\) for \(s>0\).