Large-deviation theorems for sums of random variables connected in a Markov chain. I (Q1567892)

Using the cumulant method, the authors prove large-deviation theorems for sums of random variables \(X_1,X_2,\dots,X_n\) connected in a Markov chain. To study the behavior of cumulants of \(S_n=\sum_{t=1}^nX_t\), a new scheme for enlargement of the summands \(X_t\), \(t=1,2,\dots,n\), is proposed. In this part of the paper, the following properties of the summands \(Y_1,Y_2,\dots,Y_N\) of the enlargement are proved: {A)} \({\mathbb{E}}Y_i=0\), \(i=1,2,\dots,N\); {B)} \(|{\mathbb{E}}Y_k Y_l|\leq 2(1-\beta_{k,l})^{1/2} {\mathbb{E}}^{1/2}Y_k^2{\mathbb{E}}^{1/2}Y_l^2\), \(1\leq k<l\leq N\), where \(0<\beta_{k,l}<1\) and \(1-\beta_{k,l}\leq\exp\{-(l-k-1)\}\); {C)} \({1\over {16}}\min_{1\leq t\leq n}{\mathbb{D}}X_t \leq{\mathbb{D}}Y_i \leq{36\over{\alpha^{(n)^2}}}\max_{1\leq t\leq n} {\mathbb{D}}X_t\), \(i=1,2,\dots,N\), where \(\alpha^{(n)}\) denotes the ergodicity coefficient of the Markov chain; {D)} \({1\over 24}\sum_{i=1}^N {\mathbb{D}}Y_i\leq B_n^2\leq {43\over 5}\sum_{i=1}^N {\mathbb{D}}Y_i\).