Large deviations: From empirical mean and measure to partial sums process (Q1893860): Difference between revisions

The authors show how to extend the large deviation principle from partial sums of Banach-space valued random variables to continuous-time trajectories generated by the sums. Under certain moment conditions, they show the large deviation principle for trajectories \(S_n (t) = {1 \over n} \sum_{j \leq nt} X_j\) follows from the well-studied large deviation principle for empirical measures. In particular, the large deviation principle for \(S_n (\cdot)\) is established for stationary processes that satisfy a version of hyper-mixing condition satisfied e.g. by hypercontractive Markov chains. Two applications that motivate the results are presented in the final section.