No label defined (Q4237948)

The persistence probability \(P(T)\) of a stochastic process \(X\equiv \{X(T), t\geq 0\}\) is defined as the probability that \(X\) does not change sign over the interval \([0,T]\). Let \(\{X(t), t\geq 0\}\) be a stationary Gaussian process with mean zero and \(\text{cov}(X(t), X(0))= f(t)\) with \(f(0)= 1\). The author discusses various types of stochastic processes \(X\) for which \(P(T)\simeq e^{-\Theta T}\) as \(T\to\infty\). If \(X\) is a Markov process, then \(f(t)= e^{-\lambda| t|}\) and \(P(T)\simeq e^{-\lambda T}\) as \(T\to\infty\) [cf. \textit{I. F. Blake} and \textit{W. C. Lindsey}, IEEE Trans. Inf. Theory IT-19, 295-315 (1973; Zbl 0264.60036)]. Here the author shows by heuristic arguments that, for a one-parameter family of processes \(\{X_H(t)\}\) governed by the covariance function \(f_H(t)= \text{cosh}(Ht)-{1\over 2}| 2\text{ sinh}(t/2)|^{2H}\), \(H\in (0,1)\), the persistence probability \(P(T)\simeq e^{-\theta_HT}\) as \(T\to\infty\) where \(\theta_H= 1-H\). This result is obtained by mapping the process \(X_H(t)\) to a fractional Brownian motion \(B_H(t)\) by a logarithmic time transformation and using the results for a fractional Brownian motion. Fractional Brownian motion is a nonstationary Gaussian process with stationary increments of mean zero and variance \(E| B_H(b)- B_H(S')|^2=| s-s'|^{2H}\) where \(H\in (0,1)\) is called the Hurst exponent. A plausible argument was given to show that \[ P_H(T_0, T_1)\equiv \text{Prob}[B_H(t)\neq 0,\;0< T_0< t< T_1]\simeq {T_0\choose T_1}^{(1-H)} \] for \(T_1\gg T_0\). As the author states, the paper contains no mathematical proofs and the statements made are plausible conjectures.