Continuity of eigenvalues of subordinate processes in domains (Q2580980)

Let \(E\) be a locally compact separable metric space and \(m\) a Radon measure on it. A sequence \({\mathcal E}^{n}\) of densely defined closed quadratic forms on \(L^{2}(m)\) (a quadratic form being considered to have the value \(\infty\) where it is not defined) is said to M(osco)-converge to the densely defined closed quadratic form \({\mathcal E}\) on \(L^{2}(m)\) if: \(u_{n}\in L^{2}(m)\), \(u_{n}\rightarrow u\) weakly, imply \(\liminf {\mathcal E}^{n}(u_{n},u_{n})\geq {\mathcal E}(u,u)\) and for every \(u\in L^{2}(m)\) there exist \(u_{n}\in L^{2}(m)\) with \(u_{n}\rightarrow u\) strongly and \(\limsup{\mathcal E}^{n}(u_{n},u_{n})\leq {\mathcal E}(u,u)\). If \(G_{\alpha }\) and \(P_{t}\) denote the corresponding resolvents and semigroups, then M-convergence is equivalent to \(G_{\alpha }^{n} f\rightarrow G_{\alpha } f\) in \(L^{2}(m)\) for all \(\alpha > 0\) and \(f\in L^{2}(m)\) and also to \(P_{t}^{n} f\rightarrow P_{t}f\) in \(L^{2}(m)\) for all \(t>0\) and \(f\in L^{2}(m)\). If \(E_{\lambda }\), \(\lambda \geq 0\), are the corresponding spectral families, then M-convergence implies, for every points \(\lambda <\mu\) of continuity of \(E\), \(E_{\mu }^{n}-E_{\lambda }^{n}\rightarrow E_{\mu }-E_{\lambda }\) strongly. (1) If \(E^{n}\), \(E\) have discrete spectra, then the M-convergence implies the convergence of the corresponding eigenvalues and that every limit point of eigenfunctions with norm \(1\) of \(E^{n}\) for \(\lambda^{(n)}\) is such an eigenfunction of \(E\) for \(\lambda\) if \(\lambda =\lim\lambda^{(n)}\). (2) If \(S_{n}\), \(S\) are subordinators, \(E(e^{^{-\lambda S}t})=e^{-t\Phi (\lambda )}\) for \(\lambda > 0\) and if \(\Phi_{n}\rightarrow \Phi\) pointwise on \([ 0,\infty )\), then, for every open \(D\subset E\), the forms, obtained from \({\mathcal E}\) by transforming with \(S_{n}\) and then killing at the first exit from \(D\), M-converge to the same for \(S\) and \(D\). (3) If \({\mathcal E}={\mathcal E}^{\Psi}\) corresponds to a symmetric Lévy process \(X\) on \textbf{R}\(^{d}\), \(E(e^{^{i\xi (X} t^{- X} 0^{)}})=e^{-t\Psi (\xi )}\), \(\xi \in \)\textbf{R}\(^{d}\), and if \(\Psi_{n}\rightarrow \Psi\) pointwise, then \({\mathcal E}^{^{\Psi}n}\) M-converges to \({\mathcal E}^{\Psi }\). The authors consider as their main results the application of (1) to the situations (2) and (3). They consider applications to the cases \(\Phi (\lambda ) =(\lambda +\gamma )^{\alpha /2}- \gamma^{\alpha /2}\), \(\alpha_{n}\rightarrow \alpha\), \(\gamma_{n}\rightarrow \gamma\), \(\Phi(\lambda ) =\) log(1+\(\lambda^{\alpha / 2})\), \(\alpha_{n}\rightarrow \alpha\), \(\Phi (\lambda )=((\lambda +1)^{\alpha }-1)/\alpha\), \(\alpha_{n}\rightarrow \alpha\). Possible extension to symmetric diffusion on \textbf{R}\(^{d}\) is discussed.