Second-order properties and central limit theorems for geometric functionals of Boolean models (Q259568): Difference between revisions

This paper uses the Fock space representation of Poisson functionals to explore the covariance structure of geometric functionals of \(Z\cap W\), where \(Z\) is the stationary random closed set associated to a Boolean model \(\eta\) on the space of convex bodies in \(\mathbb R^d\) and the intersection of Z with a compact and convex set \(W\subset \mathbb R^d\) is a finite union of compact convex sets. The authors also prove univariate and multivariate central limit theorems for geometric functionals of \(Z \cap W\). Section 2 recalls basic facts about stationary Poisson process \(\eta\) and its associated Boolean model \(Z\) in the Euclidean space \(\mathbb R^d\). The intensity measure of \(\eta\) is invariant under the shifts \(K\mapsto K+x\), which is equivalent to the stationarity of \(\eta\). A typical grain \(Z_0\) is a random convex set with distribution \(Q\). We have the Steiner formula \[ V_d(K+B_r^d) = \sum_{i=0}^d k_{d-i}r^{d-i}V_i(K), \quad r\geq 0, \] where the intrinsic volume \(V_i(K\cup L) = V_i(K)+V_i(L) - V_i(K\cap L)\) is additive for the Minkowski sum with radius \(r\) centered at the point \(x\in \mathbb R^d\), and \(k_n\) is the volume of the \(n\)-dimensional unit ball. The intersection \(Z\cap W\) has the well-known formula \[ \operatorname{Var} V_d(Z\cap W) = (1-p)^2\int C_W(x)(e^{\gamma C_d(x)} -1) dx, \] where \(C_W(x) = V_d(W\cap (W+x))\) is the set covariance function of a compact and convex subset \(W\), \(C_d(x)\) is the mean covariogram of the typical grain defined as \[ C_d(x) =\operatorname{E} V_d (Z_0 \cap (Z_0+x)), \quad x \in \mathbb R^d, \] and the volume fraction \(p=\operatorname{E}V_d(Z\cap [0,1]^d)\). Based on Theorem 9.1 and Theorem 4.1, a multivariate central limit theorem (Theorem 2.1) is stated in this section. That is a \((d+1)\)-dimensional central Gaussian random vector \(N\) with a covariance matrix \(\Sigma\) such that \[ \frac{1}{\sqrt{V_d(W)}} (V(Z\cap W) - \operatorname{E} V(Z\cap W)) \to N, \quad r(W) \to \infty. \] This is the first central limit theorem for the intrinsic volumes of the Boolean model beyond volume and surface area. Section 3 studies random variables of the form \(\psi (Z\cap W)\), where \(\psi \) is a real-valued measurable function defined on the convex ring. A functional \(\psi\) is geometric if it is additive, translation invariant, locally bounded and measurable. The main result of this section deals with the asymptotic behavior of the covariance between two geometric functionals of \(Z\cap W\) for expanding convex observation window \(W\) stated in Theorem 3.1. Section 4 studies the positive definiteness of the asymptotic covariance matrix for geometric functionals \(\psi_0, \dots, \psi_d\) pairwise, as considered in Section 3. Theorem 4.1 states that the covariance matrix \(\Sigma = (\sigma_{\psi_k, \psi_l})_{k, l =0, 1, \dots, d}\) is positive definite under certain assumptions like that the typical grain \(Z_0\) has nonempty interior with positive probability. Section 5 prepares some particular important intrinsic volumes formula. Theorem 5.2 shows that the measures \(H_{i, j}\) defined in this paper are all finite and \[ \int e^{\gamma C_d(x)} H_{i, j}(dx) =\rho_{i, j}, \] where \(\rho_{i, j}\) is from the asymptotic covariance of Theorem 3.1. Hence, the asymptotic covariances \[ \sigma_{i, j} = \lim_{r(W)\to \infty} \frac{\operatorname{Cov} (V_i(Z\cap W), V_j(Z\cap W))}{V_d(W)} \] can be expressed by linear combinations of the numbers \(\rho_{i, j}\) in Theorem 6.1. Section 7 shows how some of the formulas of Section 6 can be used to determine explicitly the covariances of a stationary and isotropic Boolean model. Section 8 proves a general result on the normal approximation of Poisson functionals in Theorem 8.1 in the way to prepare the central limit theorems for geometric functionals in Theorem 9.1. Under the same second moment assumptions as for the existence of the asymptotic covariances, the authors first prove the convergence in distribution and obtain rates of convergence under slightly stronger moment assumptions. For the multivariate central limit theorem, they explain that the rate is optimal. The main obstacle is that geometric functionals of \(Z\) admit an infinite chaos expansion. By bounding the kernels of the chaos expansion with monotone functionals, the authors can solve this issue. It would be interesting to see if the stabilization method can be adapted.