Uniqueness of diffusion operators and capacity estimates (Q1945688)

Let \(\Omega\subset\mathbb R^d\) be an open connected domain and let \(H=-\sum_{j,k=1}^d \partial_j c_{j,k}(x)\partial_k\) be a non-degenerate second-order differential operator in divergence form with domain \(C_c^\infty(\Omega)\). The matrix \(C=C(x)= (c_{j,k}(x))\) is assumed to be strictly elliptic, but the coefficients may have degeneracies as \(x\) approaches the boundary. Denote by \(d(x,y)\) the Riemannian metric induced by \(C^{-1}\). The operator \(H\) is said to be \(L^1\)-unique if it has a unique closed extension in \(L^1\) which generates a strongly continuous semigroup in \(L^1\); the operator is said to be Markov unique if it has a unique closed extension in \(L^2\) which generates a strongly continuous sub-Markovian semigroup in \(L^2\) (and, hence, in all spaces \(L^p\), \(1<p<\infty\)). Clearly, \(L^1\)-uniqueness entails Markov uniqueness while the converse is, in general, wrong and needs additional conditions. This paper is about the converse implication, it extends earlier papers by the author (with co-authors), where the coefficients \(c_{j,k}\) were bounded or moderately growing at infinity [\textit{D. W. Robinson} and \textit{A. Sikora}, Ann. Sc. Norm. Super. Pisa, Cl. Sci. (5) 10, No.~3, 683--710 (2011; Zbl 1259.47026); Stud. Math. 203, No.~1, 79--103 (2011; Zbl 1222.47033)]. Assume that the \(c_{j,k}\) are in the space \(W^{1,\infty}_{\text{loc}}(\Omega)\). It is shown that \(L^1\)-uniqueness and Markov uniqueness are equivalent if the Riemannian balls \(B(r,0)\) are bounded and if a Täcklind condition (i.e., \(\int^{+\infty} r [\log\text{Leb}(B(r,0))]^{-1}\,dr<\infty\) is satisfied. A similar result holds for the non-symmetric setting provided that the coefficients of the non-symmetric part satisfy certain (semi-)boundedness assumptions. The proof uses the Cauchy equation approach developed by Girgor'yan, but the geodesic completeness assumption for the underlying manifold is now replaced by the Markov uniqueness assumption. As a side-result, it is shown that under the above assumptions \(H\) is Markov-unique if, and only if, the boundary of \(\Omega\) has capacity zero. This is also equivalent to a certain approximation assumption of functions from the domain of the associated Dirichlet form (with Dirichlet boundary conditions) on each set \(A\subset\bar\Omega\).