Operator estimates for homogenization of the Robin Laplacian in a perforated domain (Q2676539)

The authors consider the perforated domain \(\Omega_{\varepsilon}=\Omega \setminus \bigcup_{i}\overline{D_{i,\varepsilon}}\), where \(\Omega\) is an open and not necessarily bounded subset of \(\mathbb{R}^{n}\) (\(n\geq 2\)) with \(C^{2}\) boundary \(\partial \Omega\), and \(D_{i,\varepsilon}\), \(i=1,\dots, \mathcal{I}(\varepsilon)\), are identical open balls of radius \(d_{\varepsilon}\) \(\varepsilon\)-periodically distributed in \(\Omega\). If \(\Omega\) is unbounded, the authors further assume that it is uniformly regular of class \(C^{2}\). They assume that \(\Lambda_{\varepsilon}=d_{\varepsilon}/\varepsilon \rightarrow 0\) when \(\varepsilon \rightarrow 0\). They consider the elliptic problem \(-\Delta u_{\varepsilon}+u_{\varepsilon}=f\in L^{2}(\Omega)\) posed in \(\Omega_{\varepsilon}\), with the homogeneous boundary conditions \(\frac{\partial u_{\varepsilon}}{ \partial n}+\gamma_{\varepsilon}u_{\varepsilon}=0\) on \(\bigcup_{i}\partial D_{i,\varepsilon}\), \(u_{\varepsilon}=0\) on \(\partial \Omega\). They quote from the literature the asymptotic behavior of the solution \(u_{\varepsilon}\) when \(\varepsilon \rightarrow 0\), which depends on the limits \(P\) of \(P_{\varepsilon}=\varkappa_{n}\frac{\gamma_{\varepsilon}d_{\varepsilon}^{n-1}}{\varepsilon ^{n}}\) and \(Q\) of \(Q_{\varepsilon}=(n-2)\varkappa_{n} \frac{d_{\varepsilon}^{n-2}}{\varepsilon ^{n}}\), if \(n\geq 3\) and \(Q_{\varepsilon}=\frac{2\pi}{\left\vert \ln d_{\varepsilon}\right\vert \varepsilon ^{2}}\), if \(n=2\), where \(\varkappa_{n}\) is the surface area of the unit sphere in \(\mathbb{R}^{n}\). In the present paper, the authors essentially prove estimates on the self-adjoint and positive operator \(\mathcal{A}_{\varepsilon}\) associated to this problem and with \(\operatorname{dom}(\mathcal{A}_{\varepsilon})\subset \operatorname{dom}(\mathfrak{a}_{\varepsilon})\), where \(\mathfrak{a}_{\varepsilon}\) is the sesquilinear form defined in the Hilbert space \(L^{2}(\Omega_{\varepsilon})\) through \(\mathfrak{a}_{\varepsilon}[u,v]=\int_{\Omega_{\varepsilon}}\nabla u\cdot \overline{\nabla v}dx+\sum_{i\in \mathcal{I}_{\varepsilon}}\gamma_{\varepsilon}\int_{\partial D_{i,\varepsilon}}u\overline{v}ds\), with \(\operatorname{dom}(\mathfrak{a}_{\varepsilon})=\{u\in H^{1}(\Omega_{\varepsilon}):u|_{\partial \Omega}=0\}\). The first main result proves that if \(P\) or \(Q\) are finite, then \[ \left\Vert (\mathcal{A}_{\varepsilon}+I)^{-1}J_{\varepsilon}-J_{\varepsilon}(\mathcal{A}+I)^{-1}\right\Vert_{L^{2}(\Omega)\rightarrow L^{2}(\Omega_{\varepsilon})}\leq C\eta_{\varepsilon}, \] where the limit operator \(\mathcal{A}\) is associated to the limit sesquilinear form \(\mathfrak{a}\) defined in the Hilbert space \(L^{2}(\Omega)\) through \(\mathfrak{a}[u,v]=\int_{\Omega}\nabla u\cdot \overline{\nabla v}dx+PQ(P+Q)^{-1}\int_{\Omega}u\overline{v}dx\), with \(\operatorname{dom}(\mathfrak{a})=H_{0}^{1}(\Omega)\), \(J_{\varepsilon}\) being the operator of restriction to \(\Omega_{\varepsilon}\) and \(\eta_{\varepsilon}\) being explicitly given in terms of \(V_{\varepsilon}-V\) (\(V_{\varepsilon} =P_{\varepsilon} Q_{\varepsilon}/(P_{\varepsilon}+Q_{\varepsilon})\)) and of \(\Lambda_{\varepsilon}\). In this case, the authors also prove that some distance between the spectra of \(\mathcal{A}_{\varepsilon}\) and \(\mathcal{A}\) is lower than \(C\eta_{\varepsilon}\). They then prove quite similar estimates for the resolvent in the cases where \(Q>0\) or \(P=Q=\infty\). For the proofs of these results, the authors mainly prove operator and improved Poincaré estimates in varying Hilbert spaces.