Upper bounds for spatial point process approximations (Q1774190): Difference between revisions

Let \(D_1\), \(D_2\in \mathbb N=\{1,2,3,...\}\) and \(D=D_1 + D_2\). Consider a point process \(\xi\) on \(\mathbb R^D=\mathbb R^{D_1}\times \mathbb R^{D_2}\), which has expectation measure \(\nu\) and meets three conditions, namely, 1) absolute continuity of \(\nu\) with a mild restriction on the density; 2) an orderliness condition in the \(\mathbb R^{D_1}\)-directions [for a detailed account of orderliness, see \textit{D. J. Daley}, in: Stochastic Geometry, Tribute Memory Rollo Davidson, 148-161 (1974; Zbl 0285.60039)]; 3) a mixing condition in the \(\mathbb R^{D_2}\)- directions. The various versions of the third condition are mixing conditions of different strength [see \textit{P. Doukhan}, ``Mixing: Properties and examples'' (1994; Zbl 0801.60027)]. Let \(\eta\) be a Poisson process with the same expectation measure and let \(\theta_T: \mathbb R^D\to \mathbb R^D\) be the linear transformation that stretches the first \(D_1\) coordinates by a factor \(\omega(T)^{1/D_1}\) and compresses the last \(D_2\) coordinates by a factor \(T^{1/D_2}\). Consider restrictions of transformed processes \(\xi \theta_T^{-1}\) and \(\eta \theta_T^{-1}\) to a bounded cube \(J=[-1, 1)^D\). It was shown by \textit{S. P. Ellis} [Adv. Appl. Probab. 18, 646--659 (1986; Zbl 0609.60059)] that, for bounded measurable functions \(f_T: J\to \mathbb R\) with \(\| f_T\| = O(\sqrt{\omega(T)/T})\), the difference between characteristic functions of \(\int_Jf_td(\xi\theta_T^{-1})\) and \(\int_Jf_td(\eta \theta_T^{-1})\) converges uniformly to zero on every compact subset of \(\mathbb R\) as \(T\to \infty\). Therefore, there is hope that \(d({\mathcal L}(\xi\theta_T^{-1}| _J), {\mathcal L}(\eta\theta_T^{-1}| _J))\) can be shown to be small for large \(T\) if we choose for \(d\) a probability distance between distributions of point processes which metrizes a topology that is equal to or not too much finer than the weak topology (i.e., the topology of convergence in distribution). It was shown by \textit{A. D. Barbour, L. Holst}, and \textit{S. Janson} [``Poisson approximation'' (1992; Zbl 0746.60002)] that \(d\) should be \(d_2\)-distance which can be defined by the formula \(d_2(P, Q) := \sup_{f\in F_2}\left| \int f\, dP - \int f\, dQ\right| \), where \(P\) and \(Q\) are probability measures on \(\mathcal M\), \(\mathcal M\) is the space of point measures on a compact set \(\mathcal X\), \(F_2\) is a certain class of functions determined on \(\mathcal M\). In the present article under the above mentioned conditions \(1)-3)\), explicit upper bounds are given for the \(d_2\)-distance between the corresponding point process distributions. A number of related results, and applications to kernel density estimation and long range dependence testing are also presented.