Markov uniqueness and essential self-adjointness of perturbed Ornstein-Uhlenbeck operators (Q1911597): Difference between revisions

Let \((E,H,m)\) be the classical Wiener space, i.e., \(E:=C(\mathbb{R}_+,\mathbb{R}^d)_0\), \(m=\)Wiener measure on \(E\), \(H:=\) corresponding reproducing kernel Hilbert space (i.e., Cameron-Martin space). Then \(E'\subset H'\equiv H\subset E\) continuously and densely. Let \(K\subset E'\) be an orthonormal basis of \(H\). Set \(\alpha_k(x):=k(x)\), \(k\in K\), \(x\in E\), and define \[ {\mathcal F}C_b^\infty(K):=\{f(\alpha_{k_1},\dots,\alpha_{k_N})\mid N\in\mathbb{N},\;k_1,\dots,k_N\in K,\;f\in C_b^\infty(\mathbb{R}^N)\}. \] Consider the corresponding Dirichlet form \(({\mathcal E},D({\mathcal E}))\) defined as the closure of \[ {\mathcal E}(u,v):=\int_E\langle\nabla u,\nabla v\rangle_Hdm,\quad u,v\in {\mathcal F}C_b^\infty(K), \] on \(L^2(E;m)\). Let \(\varphi\in D({\mathcal E})\) and consider the finite measure \(\mu:=\varphi^2m\). Define \[ Au:=\sum_{k\in K}\left({\partial^2 u\over \partial k^2}-\alpha_k{\partial u\over \partial k}\right)+2\varphi^{-1}\langle\nabla u,\nabla\varphi\rangle_H,\;\;u\in D(A):={\mathcal F}C_b^\infty(K). \] Since \(({\mathcal E},D({\mathcal E}))\) satisfies L. Gross' log-Sobolev inequality, it follows that \(\alpha_k\varphi\in L^2(E;m)\) for all \(k\in K\), and hence \(A\) is an operator on \(L^2(E;\mu)\) (i.e., \(A({\mathcal F}C_b^\infty(K))\subset L^2(E;\mu))\), which is symmetric. Let \(D(\mu)\) denote the set of all Dirichlet forms on \(L^2(E;\mu)\) whose generators extend \((A,D(A))\). Clearly, \(\#D(\mu)\geq1\). One says Markov uniqueness holds, if \(\#D(\mu)=1\). One says strong uniqueness holds, if \((A,D(A))\) is essentially selfadjoint on \(L^2(E;\mu)\). The latter clearly implies Markov uniqueness. It is well-known that indeed Markov uniqueness holds in this case. We refer to Corollary 1.6 of [\textit{S. Albeverio}, the reviewer and \textit{T.-S. Zhang}, in: Stochastic processes and optimal control. Stochastic Monogr. 7, 1-26 (1993; Zbl 0827.31007)] which even covers the case when \(m\) is replaced by a non-Gaussian measure in a certain class. It is also well-known that \((A,D(A))\) is essentially selfadjoint on \(L^2(E;\mu)\) if \(\varphi^{-1}|\nabla\varphi|_H\in L^4(E;\mu)\) [cf. \textit{V. A. Liskevich} and \textit{Yu. A. Semenov}, J. Funct. Anal. 109, No. 1, 199-213 (1992; Zbl 0788.47041)]. Surprisingly, both above results are not mentioned (and the second paper is not even included in the references) in the paper under consideration in which both results are reproved. The essential selfadjointness is even only proved under the stronger assumption that \(\varphi^{-1}|\nabla \varphi|_H\) is essentially bounded. On the other hand the method of proof is different from and to some extend simpler than the known ones, and is based on previous work by the author.