Invariance of Poisson measures under random transformations (Q1930650): Difference between revisions
From MaRDI portal
Added link to MaRDI item. |
Set profile property. |
||
Property / MaRDI profile type | |||
Property / MaRDI profile type: MaRDI publication profile / rank | |||
Normal rank |
Revision as of 06:16, 5 March 2024
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | Invariance of Poisson measures under random transformations |
scientific article |
Statements
Invariance of Poisson measures under random transformations (English)
0 references
11 January 2013
0 references
\textit{A. S. Üstünel} and \textit{M. Zakai} obtained invariance results for the Wiener measure under quasi-nilpotent random isometries in [C. R. Acad. Sci., Paris, Sér. I 319, No. 10, 1069--1073 (1994; Zbl 0810.60001)] by means of the Malliavin calculus, based on the duality between gradient and divergence operators on the Wiener space. In this paper, the author obtains sufficient conditions for the invariance of random transformations \(\tau:\Omega^X \times X\to Y\) of Poisson random measures on metric spaces \(X\) and \(Y\). For this purpose, the author use the Malliavin calculus under Poisson measures, which relies on a finite difference gradient \(D\) and a divergence operator \(\delta\) that extends the Poisson stochastic integral. The results and proofs are to some extent inspired by the treatment of the Wiener case in [\textit{A. S. Üstünel} and \textit{M. Zakai}, Probab. Theory Relat. Fields 103, No. 3, 409--429 (1995; Zbl 0832.60052)] (see, also, [the author, Electron. Commun. Probab. 14, 116--121 (2009; Zbl 1189.60113)] for a recent simplified proof on the Wiener space). More precisely, let \(X\) be a \(\sigma\)-compact metric space, and \((X, {\mathcal B}(X), \sigma)\) forms a measure space with a \(\sigma\)-finite measure \(\sigma\). Let \(\Omega^X\) denote the configuration space on \(X\). Each element \(\omega\) of \(\Omega^X\) is identified with the Radon point measure \( \omega = \sum_{i=1}^{\omega(X)} \varepsilon_{x_i}\), where \(\varepsilon_{x_i}\) denotes the Dirac measure at \(x \in X\) and \(\omega(X) \in {\mathbb N} \cup \{ \infty \}\) denotes the cardinality of \(\omega\). The Poisson random measure \(N(\omega, dx)\) is defined by \[ N(\omega, dx) = \omega(dx) = \sum_{k=1}^{\omega(X)} \varepsilon_{x_k} ( dx), \quad \omega \in \Omega^X. \] The Poisson measure \(\pi_{\sigma}\) is characterized by its Fourier transform \(\psi_{\sigma}(f)\) with expectation \(\operatorname{E_{\sigma}}\) under \(\pi_{\sigma}\): \[ \psi_{\sigma}(f) = \exp \left\{ \int_X ( e^{i f(x)} - i f(x) -1) \sigma(dx) \right\} \quad \text{for } f \in L_{\sigma}^s(X). \] The finite-difference gradient \(D\) is defined as \( D_x F(\omega):=\varepsilon_x^+ F(\omega) - F(\omega)\) for \(\omega \in \Omega^X\) and \(x \in X\), for any random variable \(F : \Omega^X \to {\mathbb R}\), where \(\varepsilon_x^+ F(\omega)=F(\omega \cup \{ x \})\) for \(\omega \in \Omega^X\) and \(x \in X\). Note that the operator \(D\) is continuous on the space \({\mathbb D}_{2,1}\) defined by the norm \[ \| F \|_{2,1}^2 := \| F \|_{ L^2( \Omega^x, \pi_{\sigma})}^2 + \| DF \|_{ L^2( \Omega^X \times X, \pi_{\sigma} \otimes \sigma )}^2 \quad \text{for } F \in {\mathbb D}_{2,1}. \] The Skorohod integral operator \(\delta_{\sigma}\) is defined on any measurable process \(u : \Omega^X \times X \to {\mathbb R}\) by the expression \[ \delta_{\sigma}(u) = \int_X u( \omega, t) ( \omega(dt) - \sigma(dt)), \] provided that \(\operatorname{E_{\sigma}}[ \int_X | u(\omega, t) | \sigma( dt) ] < \infty\). For any measurable process \(u : \Omega^X \times X \to {\mathbb R}\), let \[ \Delta_{s_0} \dotsm \delta_{s_j} \prod_{p=0}^n u_{s_p} = \sum _{\substack{ \Theta_0 \cup \dots \cup \Theta_n = \{ s_0, s_1, \dotsc, s_j \} \\ s_0 \in \Theta_0^c, \dotsc, s_j \in \Theta_j^c }} D_{\Theta_0} u_{s_0} \dotsm D_{\Theta_n} u_{ s_n}, \] for \(s_0, \dotsc, s_n \in X\) and \(0 \leqslant j \leqslant n\), where \(D_{\Theta} := \prod_{ s_j \in \Theta} D_{s_j}\) when \(\Theta\) \(\subset\) \(\{ s_0, s_1, \dotsc, s_j \}\). Let \(( Y, \mu)\) be another measure space with associated configuration space \(\Omega^Y\) and \(\sigma\)-finite intensity measure \(\mu(dy)\). Let \(\tau_*(\omega)\) \(( \omega \in \Omega^X)\) (cf. \(\tau_* : \Omega^X \to \Omega^Y\)) be the image measure of \(\omega\) by the measurable random mapping \(\tau\) : \(\Omega^X \times X \to Y\). The random isometry \(R:L_{\mu}^p(Y) \to L_{\sigma}^p(X)\) is given as \(Rh = h \circ \tau\) for \(h \in L_{\mu}^p(Y)\). Under \(\pi_{\sigma}\), \(\delta_{\sigma}(Rh)\) has the same distribution as the compensated Poisson integral \(\delta_{\mu}(h)\) of \(h\) under \(\pi_{\mu}\). Here is the main result of this paper, which indicates the invariance for Poisson measures when \((X, \sigma)=( Y, \mu)\). Theorem. Assume the cycle vanishing condition \[ D_{t_1} \tau(\omega, t_2) \dotsm D_{t_k} \tau(\omega, t_1) = 0 \quad \forall \omega \in \Omega^X\;\forall t_1, \dotsc, t_k \in X \] for all \(k \geq 1\). Then \(\tau_* : \Omega^X \to \Omega^Y\) maps \(\pi_{\sigma}\) to \(\pi_{\mu}\), i.e., \(\tau_* \pi_{\sigma} = \pi_{\mu}\) is the Poisson measure with intensity \(\mu(dy)\) on \(Y\).
0 references
Poisson measure
0 references
random transformation
0 references
invariance
0 references
Skorohod integral
0 references
moment identities
0 references