Estimating the distribution of one-dimensional discrete scan statistics viewed as extremes of 1-dependent stationary sequences (Q866625)

Let \(X_1, X_2,\dots, X_n\) be a sequence of i.i.d. (independent and identically distributed) Bernoulli \(B(1, p)\) random variables, \(P(X_1 = 1) = p = 1 - P(X_1 = 0)\). Let \(1 \leq m \leq n\) be a fixed positive integer, and let \(Y_t =\sum _{i=t}^{t+m-1}X_i\), \(1 \leq t\leq n - m + 1\). The one-dimensional discrete scan statistic generated by \(\{X_n\}\) is defined as \(S = S_m(n) =\max_{1\leq t\leq n-m+1} Y_t\). Various approximation methods and bounds for \(P(S \leq k)\) have been proposed, and the author [Extremes 3, No. 4, 349--361 (2000; Zbl 1003.60075)] obtained approximations of the distributions of extremes for one-independent stationary sequences together with sharp bounds for the corresponding errors. The aim of this paper is to present a method that provides accurate approximations for \(P(S \leq k)\), with sharp bounds for the corresponding errors, which is particularly suitable for large \(m\) and \(n / m\). The author and \textit{C. Preda} [Methodol. Comput. Appl. 4, No. 4, 393--407 (2002; Zbl 1031.62066); ibid. 8, No. 3, 373--382 (2006; Zbl 1108.62099)] have obtained improved estimations for the distribution of one-dimensional continuous scan statistics generated by a Poisson process, and for the distributions of two-dimensional scan statistics. All these results are then applied to discrete one-dimensional scan statistics generated by Bernoulli random variables, including to the longest success run and to the charge problem.