An almost-Markov-type mixing condition and large deviations for Boolean models on the line (Q996736)

The Boolean model also called Poisson grain model is one of the best studied and mostly used random set models to describe irregularly shaped and randomly scattered clumps in Euclidean spaces \(\mathbb R^d\) of arbitrary dimension \(d\geq1\). A non-stationary Boolean Model in \(\mathbb R^d\) is the union set \(\Xi= \bigcup_{i\geq1}(\Xi_i + X_i)\) of independent copies \(\Xi_1,\Xi_2,\dots\) (called grains) of a random compact set \(\Xi_0\sim P_0\) (typical grain) having the distribution \(P_0\), where the grains are independently shifted by the atoms \(\Xi_1,\Xi_2,\dots\) (called germs) of a non-stationary Poisson point process \(\Psi=\sum_{i\geq1}\delta_{X_i}\sim \Pi_{\Lambda}\) driven by the intensity measure \(\Lambda\). Here \(\delta_x(B)=1(=0)\), if \(x\in B\) (\(x\not\in B\)) and \(\Lambda(B)=E\Psi(B)\) (=mean number of germs in \(B\)) for any bounded Borel set \(B\subset \mathbb R^d\) (\(E\) stands for expectation). Note that the random set \(\Xi\) is a.s. closed, if \(E\Lambda((-\Xi_0)+ K)<\infty\) for all compact sets \(K\subset\mathbb R^d\). The distribution of the Boolean model \(\Xi\) is determined by its hitting functional \(T_{\Xi}(K)=1-\exp\{-E((-\Xi_0)+K)\}\) (\(K\) is compact). The present paper considers Boolean models on the real axis \(\mathbb R^1\). The particular case, where \(\Xi_0\) is a random line segment \([0, X_0]\), is considered in queuing theory in connection with the infinite server queue \(M/G/\infty\). Further, stationary Boolean models in \(\mathbb R^1\) occur in stochastic geometry and stereology to analyze higher-dimensional Boolean models via linear sections. Stationary one-dimensional Boolean models with typical grain \(\Xi_0=[0, X_0]\) can be modeled by an alternating renewal process, where the two phases correspond to spacings and clumps, respectively. The author has used the following two basic assumptions: (BM1) The intensity measure \(\Lambda(\cdot)\) possesses a bounded Lebesgue density \(\lambda(t)\), \(t\in \mathbb R^1\), that is, there exists some \(\lambda_0\in (0,\infty)\) such that \(\Lambda(B) = \int_B\lambda(t) \,dt\leq \lambda_0 | B| \) for all bounded Borel sets \(B\subset \mathbb R^1\). (BM2) The typical grain \(\Xi_0\) is a random compact set in \(\mathbb R^1\) with diameter \(\| \Xi_0\| =\sup\{x-y: x,y\in\Xi_0\}\) satisfying the condition \(M_a=E\exp\{a \| \Xi_0\| \}<\infty\) for some \(a>0\). Theorem 1 of the paper states that the conditions (BM1) and (BM2) imply the analyticity of the cumulant-generating function \(L_T(z)= \log E\exp\{z | \Xi\cap[0,T]| \}\) on some open disc \(D_{\varepsilon}=\{z\in C^1: | z| <\varepsilon\}\) in the complex plane, where its radius \(\varepsilon=\varepsilon(a, \lambda_0)>0\) does not depend on \(T\in(0,\infty)\). This fact can be expressed in terms of the cumulants (the \(k\)th-order mixed cumulant of a random variable \(Y\) is defined by \(\text{Cum}_k(Y)=i^k {{d^k}\over{ds^k}} \log E\exp\{isY\}| _{s=0}\)) of the total length \(| \Xi\cap [0, T]| \) of the Boolean model \(\Xi\) in the \([0, T]\). The main results of the paper (Theorems 1 and 2) are related with some incompletely proved results in [\textit{V. A. Statulevičius}, Litov. Mat. Sb. 10, 583--592 (1970; Zbl 0266.60014)]. Theorem 1. Let the one-dimensional Boolean model \(\Xi\) satisfy the assumptions (BM1) and (BM2). Then for any \(k\geq2\) and \(T>0\), \(| \text{Cum}_k\{| \Xi\cap[0,T]| \}| \leq T H(a) \Delta(a)^{k-2} k!\), where \(H(a)=400 e \lambda_0 M_a/a^2\). Furthermore, if additionally \(\lambda(t)\equiv \lambda_0\), then the (thermodynamic) limit \(L(z)=\lim_{T\to \infty}L_T(z)/T\) exists and is an analytic function on the open disc \(D_{\varepsilon}\) with radius \(\varepsilon=1/\Delta(a)\). Theorem 2 of the paper is derived from the estimate in Theorem 1 combined with a well-known lemma on large deviations for a single random variable [see Lemma 2.3 from \textit{L. Saulis} and \textit{V. A. Statulevičius}, Limit theorems for large deviations. Dordrecht etc.: Kluwer Academic Publishers (1991; Zbl 0744.60028)].