Irregularity of the Bergman projection on worm domains in \(\mathbb C^{n}\) (Q1940080)

Let \(\Omega\) denote a bounded domain in \(\mathbb C^n\). By \(A^2(\Omega)\) we denote the Bergman space of square-integrable holomorphic functions on \(\Omega\). There exists a hermitian orthogonal projection mapping \(P_\Omega :L^2(\Omega)\longrightarrow A^2(\Omega)\), which is known as the Bergman projector. In the article under review the question whether \(P_\Omega\) maps the \(L^p\)-Sobolev space \(W^{p,s}(\Omega)\) of order \(s\) into itself or not is studied for the case that \(\Omega\) belongs to a series of smooth bounded pseudoconvex domains \(\Omega_{\alpha\beta}\) in \(\mathbb{C}^n\) that arise from a generalization of the Diederich-Fornæss worm domains in \(\mathbb C^2\). \smallskip The domains \(\Omega_{\alpha\beta}\) are defined as follows: For some \(M>0\) let \(\sigma (t):=Me^{-1/t}\) for \(t>0\) and \(\sigma (t):=0\) for \(t \leq 0\). Then, with \(\alpha>0\), \(\beta>1\) let (for \(n \geq 3\)) \[ \Omega_{\alpha\beta}:=\big\{z \in \mathbb{C}^n |\,r_{\alpha\beta}<0\big\} \,, \] where \[ r_{\alpha\beta} (z):= \Big|z_1-e^{2i\alpha\,\ln \,|z_n|}\Big|^2 +|z'|^2-1 +\sigma(|z_n|^2-\beta^2)+\sigma(1-|z_n|^2) \] and \(|z'|^2:=|z_2|^2+...+|z_{n-1}|^2\). The authors show that \(\Omega_{\alpha\,\beta}\) is pseudoconvex if \(M\) is sufficiently large and then establish the following result on lack of regularity of \(P_\Omega\), namely \smallskip Theorem. Let \(\Omega:=\Omega_{\alpha\,\beta}\). Then \(P_\Omega\) does not map \(W^{p,s}(\Omega)\) to \(W^{p,s}(\Omega)\) when \[ p \in [1,\infty)\quad\text{and}\quad s \geq \frac{\pi}{2\alpha \,\ln \beta}+ n \Big(\frac{1}{p}-\frac{1}{2}\Big). \] \smallskip For \(p=2\) it follows immediately that \(P_\Omega\) does not map \(W^{2,s}(\Omega)\) to \(W^{2,s}(\Omega)\) if \(s \geq \frac{\pi}{2\alpha \,\ln \beta}\). \smallskip In the case that \(s=0\), the authors obtain the Corollary. If \(p\) satisfies at least one of the conditions \[ 0<\frac{1}{p} \leq \frac{1}{2}- \frac{\pi}{2n\,\alpha \,\ln \beta}\qquad \text{or} \qquad \frac{1}{2}+ \frac{\pi}{2n\,\alpha \,\ln \beta} \leq \frac{1}{p} <1, \] then \(P_\Omega\) does not map \(L^p(\Omega)\) to \(L^p(\Omega)\).