Square-Gaussian stochastic processes (Q2732668)

This article may be considered as the second part of the article by the same authors [Extremes 2, No. 3, 269-293 (1999; Zbl 0961.60043)]. The authors consider the square-Gaussian stochastic processes. A square-Gaussian process \(\eta(t)\), in general, is a process that may be represented in the form NEWLINE\[NEWLINE\eta(t)=\vec{X}^{\top}(t)A(t)\vec X(t)-E\vec{X}^{\top}(t)A(t)\vec{X}(t),NEWLINE\]NEWLINE where \(\vec{X}(t)\), \(t\in T\), is a Gaussian vector-valued process and \(A(t)\) is a matrix, or \(\eta(t)\) is a mean square limit of processes represented in such a form. Note that square-Gaussian stochastic processes belong to a wide class of stochastic processes which satisfy the Cramér condition. A process which satisfies the Cramér condition may be considered as a process from the Orlicz space generated by the \(C\)-function \(\eta(x)=\exp\{|x|\}-1\) [see \textit{Yu. V. Kozachenko}, Theory Probab. Math. Stat. 30, 103-117 (1985); translation from Teor. Veroyatn. Mat. Stat. 30, 92-107 (1984; Zbl 0571.60053) and ibid. 31, 51-58 (1985) resp. ibid. 31, 44-50 (1984; Zbl 0593.60048)] or by the \(N\)-function \(U(x)=\exp\{|x|\}-|x|-1\) [see \textit{M. A. Krasnosel'skij} and \textit{Ya. B. Rutitskij}, ``Convex functions and Orlicz spaces'' (1958; Zbl 0084.10104)]. Centered processes from such Orlicz space are called pre-Gaussian stochastic processes. These processes were introduced in different form in the paper by \textit{V. V. Buldygin} and \textit{Yu. V. Kozachenko} [ibid. 10, 37-45 (1976), resp. ibid. 10, 39-47 (1974; Zbl 0299.60030)]. NEWLINENEWLINENEWLINEMany authors investigated properties of the distribution of suprema of stochastic processes. They investigated the problem of existence of moments and exponential moments of the distribution of suprema of processes, estimates of probability \(P\{\sup_{t\in T}|X(t)|>\varepsilon\}\), distribution of the number of exceedings of a certain level, etc. The classical results may be found in books and papers by \textit{H. Cramér} and \textit{M. R. Leadbetter} [``Stationary and related stochastic processes. Sample function properties and their applications'' (1967; Zbl 0162.21102)], \textit{M. B. Marcus} and \textit{G. Pisier} [``Random Fourier series with applications to harmonic analysis'' (1981; Zbl 0474.43004)], \textit{M. R. Leadbetter, G. Lindgren} and \textit{H. Rootzén} [``Extremes and related properties of random sequences and processes'' (1983; Zbl 0518.60021)], \textit{M. Ledoux} and \textit{M. Talagrand} [``Probability in Banach spaces. Isoperimetry and processes'' (1991; Zbl 0748.60004)], \textit{M. Talagrand} [Ann. Probab. 24, No. 3, 1049-1103 (1996; Zbl 0867.60017)], \textit{V. V. Buldygin} and \textit{Yu. V. Kozachenko} [``Metric characterization of random variables and random processes'' (1998; Zbl 0933.60031)], \textit{Yu. V. Kozachenko} and \textit{O. M. Moklyachuk} [Theory Probab. Math. Stat. 50, 89-98 (1995); translation from Teor. Jmovirn. Mat. Stat. 50, 87-96 (1994; Zbl 0861.60027)], \textit{Yu. V. Kozachenko} and \textit{T. A. Oleshko} [ibid. 47, 57-64 (1993), resp. ibid. 47, 55-62 (1992; Zbl 0842.60039)] and \textit{Yu. V. Kozachenko} and \textit{O. V. Stus} [Math. Commun. 3, No. 1, 83-94 (1998; Zbl 0910.60021)]. NEWLINENEWLINENEWLINEIn this paper the authors propose estimates of large deviation probability for the class of square-Gaussian stochastic processes and some subclasses of this class. These subclasses are determined by a function \(R(s)\), \(-a_1<s<a_2,\) such that NEWLINE\[NEWLINE \sup_{t\in T}E\exp\left\{ {{sX(t)}\over{\left(\text{Var} X(t)\right)^{1/2}}} \right\}\leq R(s). NEWLINE\]NEWLINE The estimates are determined by the function \(R(s)\) and the function \(\sigma(h)\), \(h>0,\) such that NEWLINE\[NEWLINE\sup_{\rho(t,s)\leq h} \left(E\left(X(t)-X(s)\right)^2\right)^{1/2} \leq\sigma(h). NEWLINE\]NEWLINE Estimates are more precise for more restricted classes of stochastic processes.