Extended large deviations (Q1332408)

Consider an i.i.d. sequence \(X_ 1, X_ 2, \dots\) of real-valued random variables having a finite moment generating function \(\Phi (t)\) in a neighborhood of zero. Let \(S_ n = X_ 1 + \cdots + X_ n\), \(S_ 0 = 0\), denote their partial sums. The authors study an extended large deviation theorem, i.e. they discuss convergence of \((nc_ n)^{-1} \log P (S_ n > n \gamma_ n)\), whenever \(\gamma_ n \to \infty\), and where \(c_ n\) is a sequence related to \(\gamma_ n\) via the Chernoff function \(h(x) = \sup_ t \{ct - \log \Phi (t)\}\). The latter result is used to establish an extended functional large deviation principle complementing earlier results of \textit{S. R. S. Varadhan} [Commun. Pure Appl. Math. 20, 431-455 (1967; Zbl 0155.165)] and \textit{J. Lynch} and \textit{J. Sethuraman} [Ann. Probab. 15, 610-627 (1987; Zbl 0624.60045)]. They also derive an extended functional Erdős-Rényi law dealing with increments of the partial sum process over subintervals of size \(k_ n = o(\log n)\). This is an interesting supplement of the functional Erdős- Rényi laws due to \textit{K. A. Borovkov} [Theory Probab. Appl. 35, No. 4, 762-766 (1991); translation from Teor. Veroyatn. Primen. 35, No. 4, 758- 762 (1990; Zbl 0717.60042)] and \textit{P. Deheuvels} [Stud. Sci. Math. Hung. 26, No. 2/3, 261-295 (1991; Zbl 0767.60029)] who considered the case of \(k_ n \sim c \log n\) for \(0 < c < \infty\) under different regularity conditions.