Iteration of the lent particle method for existence of smooth densities of Poisson functionals (Q1935425): Difference between revisions

The authors consider a local symmetric Dirichlet structure \({\mathcal S}= (\Xi ,X,\nu ,d,\gamma )\) (defined by \textit{N. Bouleau} and \textit{F. Hirsch} [Dirichlet forms and analysis on Wiener space. Berlin etc.: de Gruyter (1991; Zbl 0748.60046)]), in which \((\Xi ,X,\nu )\) is a \(\sigma\)-finite measure space, \(\{x\}\in X\), \(\nu (\{x\})=0\), \(\gamma\) is the carré du champs operator and \(d\) the domain (supposed separable), \(e[f,g]=\int \gamma [f,g]d\nu /2\) is a local Dirichlet form with domain \(d\). Also, a probability space \((R,{\mathcal R},\varrho )\) with infinite-dimensional \(L^{2}\), the gradient operator \(u\rightarrow u^{\flat }\) with values in \(L_{0}=\{g;<g,1>=0\}\subset L^{2}(\varrho )\), the generator \((a,D(a))\), under the hypothesis (H): There exists a subspace \(H\subset D(a)\cap L^{1}(\nu )\) dense in \(L^{1}(\nu )\cap L^{2}(\nu )\) with \(\gamma [f]\in L^{2}(\nu )\) for \(f\in H\). And two probability space \(\Omega _{1}\), \(\Omega _{2}\), their product \((\Omega ,P)\), a \(dt\otimes \nu\) Poisson random measure on \(\Omega _{1}\), \(N^{\sim }=N-\nu \otimes dt\), the Poisson random measures \(N\odot \varrho ^{\odot n}\), \(n\geq 1\), on a \((\Omega \times \Omega ',P\otimes P')\), constructed, if \(N=\sum _{i}\varepsilon _{(\alpha _{i},u_{i})}\), as \(N\odot \varrho ^{\odot n}=\sum _{i}\varepsilon _{(\alpha _{i},u_{i},r_{i}^{1},\dots,r_{i}^{n})}\), with the \(r\)'s, on \(\Omega '\), being independent and \(\varrho \) distributed. Let \(d_{0}\) be the linear hull of \(e^{iN^{\sim }(f)}\), \(f\in (H\otimes 'L^{2}(dt))\cap L^{1}(\nu \otimes dt)\) (\(\otimes '\) being the algebraic tensor product), \(A_{0}e^{iN^{\sim }(f)}=e^{iN^{\sim }(f)}(iN^{\sim }(a[f])-N(\gamma [f])/2)\), the actions of \(a\) and \(\gamma\) being the corresponding on the product of \({\mathcal S}\) with the trivial structure on \(L^{2}(dt)\). Consider the closed Hermitian \(({\mathcal E},{\mathcal D})\) generated by the Friedrichs extension \(A\) of \(A_{0}\), \(\Gamma\) its carré de champ and \(\sharp \) its gradient. Hypotheses C: There exists a subspace \(d_{0}\subset \cap _{p\geq 2}L^{p}(\nu )\), dense in \(d\), such that, \(n\flat \) being the iterate of \(\flat \), \(u^{n\flat }\in d(L_{0}^{\otimes n})\cap \cap _{p\geq 2}L^{p}(\nu ,L_{0}^{\otimes n})\) for all \(n\geq 1\). Let \(d^{n,p}\) be the completion of \(d_{0}\) with respect to \(\sum_{k=0}^{n}\| u^{k\flat }\| _{L^{p}(\nu ,L_{0}^{\otimes k})}\), \(d^{\infty }=\cap _{p\geq 2,n}d^{n,p}\), \({\mathcal D}_{0}\) be the set of all \(\varphi (N^{\sim }(f_{1}),\dots,N^{\sim }(f_{k}))\) with \(\varphi \in C_{c}^{\infty }(\mathbb{R}^{k})\), \(f_{i}\in L^{2}(dt)\otimes 'd^{\infty }\), \({\mathcal D}^{n,p}\) the completion of \({\mathcal D}_{0}\) for \(\sum _{k=0 }^{n}\| u^{k\sharp }\| _{L^{p}(P,L^{2}(P'{}^{\otimes n}))}\) and \({\mathcal D}^{\infty }=\cap _{p\geq 2,n}{\mathcal D}^{n,p}\). Let also \({\overline{\mathcal D}}^{\infty }= \{\)X\(\in {\mathcal D}^{\infty }\cap d(A),AX\in {\mathcal D}^{\infty }\}\). The central result is Proposition 16: If \(X\in ({\overline{\mathcal D}}^{\infty })^{d}\) and \(\Gamma (X)^{-1}\in \cap _{p\geq 2}L^{p}(P, \mathbb{R}^{d\times d})\) then \(X\) admits a density in \(C^{\infty }(\mathbb{R}^{d})\). This proposition is applied to the Poisson driven SDE \(X_{t }= x_{0}\)+ \(\int _{0 }^{t}\int c(s,X_{s-},u)N^{\sim }(dsdu)+\int _{0}^{t}\sigma (s,X_{s-})dZ_{s}\), where \(Z\) is a \(\mathbb{R}^{n}\)-valued semimartingale on \(\Omega _{2}\) with \(E((\int _{0}^{t}h_{s}dZ_{s})^{2})\leq C E(\int _{0}^{t}|h_{s}|^{2}ds)\), the coefficients being predictable, satisfying conditions ensuring the existence and unicity of the solution. One of these conditions is inforced to: \(d_{x }^{\alpha }c (t,x,\cdot\,)\in \overline {d}^{\infty }\) and \(\sup_{t\in [0,T]}\| D_{x}^{\alpha }c(t,x,\cdot )\| _{\overline {d}^{n,p}}\in \cap _{p\geq 2}L^{p}(\Omega _{2})\), where \(\overline {d}^{\infty }= \cap _{p\geq 2 ,n}\overline {d}^{n,p}\), \(\overline {d}^{n,p}\) is the completion of \(d^{\infty }\) for \(\| u\| _{n,p}+\| a(u)\| _{n,p}\). Theorem 26: If \(\Gamma (X_{t})^{-1}\in \cap _{p\geq 2}L^{p}(P,\mathbb{R}^{d\times d})\) then \(X_{t}\) has a density in \(C^{\infty }(\mathbb{R}^{d})\). \(\Gamma (X_{t})\) is expressed as an integral in terms of \(K_{s}\), \((K_{s})^{-1}\), these being solutions of other SDEs. The authors use the Dirichlet structure \({\mathcal D}\otimes d\), with \(\Gamma [f(\,\cdot ,x)](w)+\gamma [f(\omega ,\cdot )](x)\) and \((f^{\sharp }(\,\cdot ,x)(w),f^{\flat }(w,\cdot )(x))\) as \(\gamma \) and \(\flat \). The paper finishes with a paragraph of applications: regular case, nonlinear subordination (\({\mathcal S}=[0,\infty )\times W\), \(\nu =\tau \otimes m\), \((W,m)\) Wiener, \(\tau\) the infinite Lévy measure of a subordinator, \(d\) the product of the trivial one on \(L^{2}(dt)\) and of that associated with the Uhlenbeck Ornstein operator), nonlinear subordination of a diffusion, diffusive particle subjected to a Lévy field of force. There is also a paragraph discussing the classical cases, in which the equivalence of the norms \(\Gamma _{k }(F)^{1/2}\) and \(\| F^{k\sharp }\| _{p}\), \(F\in {\mathcal D}^{k,p}\), and also of \(\| (-A)^{1/2}F\|\) and\(\| \Gamma (F)^{1/2}\| _{p}\), \(F\in {\mathcal D}\), (and another such result) are established.