How big are the lag increments of a Gaussian process? (Q1591957): Difference between revisions
From MaRDI portal
Latest revision as of 11:48, 30 July 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | How big are the lag increments of a Gaussian process? |
scientific article |
Statements
How big are the lag increments of a Gaussian process? (English)
0 references
14 January 2001
0 references
Since 1983 several results on the limit behaviour of increments of Wiener processes have been investigated which may be called ``LIL (law of the iterated logarithm) generalizations''. The authors of this paper have chosen the following result of \textit{G. Chen}, \textit{F. Kong} and \textit{Z. Lin} [Ann. Probab. 14, 1252-1261 (1986; Zbl 0613.60027)] to generalize further: Let \(\{W(t), 0\leq t\leq \infty\}\) be a Wiener process and \(W(s, t)= W(s)- W(s- t)\), \(s\geq t\geq 0\). Then almost surely \[ \limsup_{T\to\infty} \sup_{0\leq t\leq T} D^{-1}|W(T,t)|= 1, \] \[ \lim_{T\to\infty} \sup_{0\leq t\leq T} \sup_{0\leq s\leq t} D^{-1}|W(T, s)|= 1,\quad \lim_{T\to \infty} \sup_{0\leq t\leq T} \sup_{t\leq s\leq T} D^{-1}|W(s, t)|= 1, \] where \(D= D(T, t)= \{2t(\log(T/t)+ \log\log t)\}^{1/2}\), \(\log t= \ln(t\vee 1)\). There are two main results in the paper which extend this statement to the general form of an almost surely continuous, centered Gaussian process \(X(t)\) with stationary increments \(E(X(t)- X(s))^2= \sigma(|t-s|)\), where \(\sigma(\cdot)\) is some special nondecreasing function, generalizing the case of a Wiener process (when \(\sigma(t)= \sqrt t\)). The first one is very similar to that mentioned above, because one has only to substitute \(W\) by \(X\) and \(t\) by \(\sigma^2(t)\) (we mean \(D\) by \(d= d(T, t)= \{2\sigma^2(t)(\log(T/t)+ \log\log t)\}^{1/2}\)). These two results are established in some conditions on \(\sigma^2(t)\) (concavity, twice-differentiability, and so on) and use the technique of estimating upper bounds of large deviation probabilities on suprema of the Gaussian process.
0 references
0 references