Symmetric reflected diffusions (Q1316640): Difference between revisions

Let \(D\) be a domain in \(\mathbb{R}^ d\), \(a\) a bounded symmetric, locally elliptic, locally Lipschitz continuous \(d \times d\) matrix-valued function on \(D\), and \(p\) a strictly positive locally Lipschitz continuous probability density on \(D\). Consider the symmetric Dirichlet form \[ {\mathcal E} (f,f):={1 \over 2} \int \nabla f \cdot a \nabla f p dx,\quad f \in D ({\mathcal E}),\;D({\mathcal E}):=\{f \in L^ 2(D;pdx) \cap H^ 1_{\text{loc}} (D) \mid {\mathcal E} (f,f)<\infty\} \] on \(L^ 2 (D;pdx)\). Let \(X\) be the stationary, symmetric Markov process associated with \(({\mathcal E}, D({\mathcal E}))\), which under more smoothness assumption is a diffusion on \(D\) with generator \({1\over 2p}\nabla\cdot(ap\nabla)\) and conormal reflection at the boundary of \(D\). In this very interesting paper two methods of approximating \(X\) by ``nicer'' diffusions are presented. In both cases tightness is established using the Lyons-Zheng decomposition [cf. \textit{T. J. Lyons} and \textit{W. Zheng}, Les processus stochastiques, Coll. Paul Lévy, Palaiseau/Fr. 1987, Astérisque 157- 158, 249-271 (1988; Zbl 0654.60059)]. Also conditions ensuring that \(X\) is a semimartingale are derived.