Local conditioning in Dawson-Watanabe superprocesses (Q1942120)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Local conditioning in Dawson-Watanabe superprocesses |
scientific article |
Statements
Local conditioning in Dawson-Watanabe superprocesses (English)
0 references
15 March 2013
0 references
This paper treats approximation theorems for a locally finite Dawson-Watanabe superprocess (DW-process) by local conditioning techniques. Some of the results are based upon the author's previous work [Ann. Probab. 36, No. 6, 2176--2214 (2008; Zbl 1167.60010)], in which he considered the local properties of a DW-process at a fixed time \(t > 0\). In particular, local finiteness of the random measure \(\xi_t\) is technically due to a precise criterion (Lemma 3.2) in that paper. Let \(\xi = ( \xi_t )\) be a DW-process in \({\mathbb R}^d\), and \(\operatorname{P}_{\mu}\) denotes probabilities under the assumption \(\xi_0 = \mu\) for the initial measure. The associated distributions are denoted by \({\mathcal L}_{\mu}(\xi)\) or \({\mathcal L}_{\mu}( \xi_t )\). Conditional probabilities and distributions are written as \(\operatorname{P}[ \cdot | \cdot ]\) and \({\mathcal L}[ \cdot | \cdot ]\), Palm measures and distributions as \(\operatorname{P}[ \cdot \| \cdot ]\) and \({\mathcal L}[ \cdot \| \cdot ]\), respectively. A random measure \(\xi\) on a measurable space \(S\) is defined as a kernel from the basic probability space \(\Omega\) into \(S\). For any random element \(\eta\) in a measurable space \(T\), \(\operatorname{P}[ \cdot \| \xi]\) may be regarded as a kernel from \(S\) to the basic probability space \(\Omega\) with \(\sigma\)-field generated by \(\eta\). The multivariate Palm distributions are defined as the kernels \({\mathcal L}[ \eta \| \xi^{ \otimes n} ]\) from \(S^n\) to \(T\), for arbitrary \(n \in {\mathbb N}\). Let \(p_t\) denote the continuous density of the symmetric Gaussian distribution on \({\mathbb R}^d\) with variance \(t >0\). The author studies the pseudo-random measures \(\tilde{\xi}\) and \(\tilde{\eta}\), which provide local approximations of the DW-process \(\xi\) in \({\mathbb R}^d\) and its canonical cluster \(\eta\). When \({\mathcal L}_{\mu}[ \xi_t \| \xi_t ]_x\) (resp., \({\mathcal L}_{\mu}[ \eta_t \| \eta_t ]_x\)) denotes the continuous version of the Palm distribution, we write \({\mathcal L}_{\mu}^0(\xi_t)={\mathcal L}_{\mu}[ \xi_t \| \xi_t ]_0\) (resp., \({\mathcal L}_{\mu}^0( \eta_t)={\mathcal L}_{\mu}[ \eta_t \| \eta_t ]_0\) ). For any space \({\mathcal M}_S\) of locally finite measures on a measure space \(S\) and any measurable set \(B \subset S\), we write \(\| \cdot \|_B=\| \cdot \|_{H_B}\) as the total variation on \(B\), where \(H_B=\{ \mu \in {\mathcal M}_S ; \mu B > 0 \}\). For a DW-process \(\xi\) in \({\mathbb R}^d\) \(( d \geq 3)\) with canonical cluster \(\eta\), there exists a pseudo-random measure \(\tilde{\xi}\) on \({\mathbb R}^d\) such that \[ \| {\mathcal L}_{\mu}( \xi_t) - \mu p_t {\mathcal L}( \tilde{\xi}) |_{B_0^{\varepsilon} } \ll \varepsilon^{d-2} \quad \text{and} \quad \| {\mathcal L}_{\mu}^0( \xi_t) - {\mathcal L}^0( \tilde{\xi}) \|_{B_0^{\varepsilon}} \to 0 \qquad (\varepsilon \to 0) \] for fixed \(\mu\) and \(t >0\), where we write \(f \ll g\) for \(f/g \to 0\), \(\tilde{\xi}\) is stationary with \(\operatorname{E} \tilde{\xi}=\lambda^{\otimes d}\) and \({\mathcal L}( \tilde{\xi})\) is an invariant measure for \(\xi\). Moreover, similar results hold for the pseudo-random measure \(\tilde{\eta}\) with \(\xi_t\) replaced by \(\eta_t\). Let \(q_{\mu,t}\) denote the continuous versions of the moments densities of the moment measure \(E_{\mu} \xi_t^{\otimes n}\). Define \[ c_d = \lim_{\varepsilon \to 0} \varepsilon^{2-d} \operatorname{P}_0 \{ \xi_t B_0^{\varepsilon} > 0 \} / p_t(0), \quad \text{for } d \geq 3 \] and \[ m_{\varepsilon} = | \log \varepsilon | \cdot \operatorname{P}_{\lambda^2} \{ \eta_1 B_0^{\varepsilon} > 0 \}, \quad \text{for }d=2. \] The main theorem reads as follows. Theorem. Let \(\varepsilon \to 0\) for fixed \(\mu\), \(t > 0\) and open \(G \subset {\mathbb R}^d\) \((d \geq 2)\). Then (i) for any \(x \in ( {\mathbb R}^d)^{(n)}\), \[ \operatorname{P}_{\mu} \{ \xi_t^n B_x^{\varepsilon} > 0 \} \sim q_{\mu,t} (x) \left\{ \begin{aligned} c_d^n \varepsilon^{n ( d -2)}, \quad & d \geq 3 \\ m_{\varepsilon}^n | \log \varepsilon |^{-n}, \quad & d = 2; \end{aligned} \right. \] (ii) for any \(x \in G^{(n)}\), in total variation on \(( B_0^1)^n \times G^c\), \[ {\mathcal L}_{\mu} [ ( \xi_t S_{x_j}^{\varepsilon} )_{j \leqslant n}, \xi_t | \xi_t^{\otimes n} B_x^{\varepsilon} > 0 ] \approx {\mathcal L}^{\otimes n} [ \tilde{\eta} S_{\varepsilon} | \tilde{\eta} B_0^{\varepsilon} > 0 ] \otimes {\mathcal L}_{\mu} [ \xi_t \| \xi_t^{ \otimes n} ]_x; \] (iii) for \(d \geq 3\), in the same sense, \[ {\mathcal L}_{\mu} [ ( \varepsilon^{-2} \xi_t S_{x_j}^{\varepsilon} )_{j \leqslant n}, \xi_t | \xi_t^{\otimes n} B_x^{\varepsilon} > 0 ] \to {\mathcal L}^{\otimes n} [ \tilde{\xi} | \tilde{\xi} B_0^1 > 0 ] \otimes {\mathcal L}_{\mu} [ \xi_t \| \xi_t^{ \otimes n} ]_x, \] where we write \(f \sim g\) for \(f/g \to 1\) and \(f \approx g\) for \(f-g \to 0\). Here, (i) extends some asymptotic results for \(n=1\) by, e.g., \textit{D. A. Dawson} et al. [Probab.\ Theory Relat. Fields 83, No. 1, 135--205 (1989; Zbl 0692.60063)]. Parts (ii) and (iii) show that, asymptotically as \(\varepsilon \to 0\), the contributions of \(\xi_t\) to the sets \(B_{x_1}^{\varepsilon},\dotsc, B_{x_n}^{\varepsilon}\) and \(G^c\) are conditionally independent. This main result includes some recursive formulas for the moment measures of \(\xi\), with connections to the uniform Brownian tree, a Brownian snake representation of Palm measures, continuity properties of conditional moment densities. The proofs are based upon the Cox cluster representations of the historical process and involve some delicate estimates of moment densities.
0 references
measure-valued branching diffusion
0 references
moment measure
0 references
Palm distribution
0 references
local approximation
0 references
local conditioning
0 references
historical process
0 references
cluster representation
0 references
Brownian snake
0 references
0 references
0 references