The Dirichlet-to-Neumann operator via hidden compactness (Q2452468)

In this very interesting article, form methods are developed to investigate generalized Dirichlet-to-Neumann operators on \(L_2(\partial\Omega)\) for a bounded Lipschitz domain \(\Omega\subset\mathbb{R}^d\). Denoting the Dirichlet Laplacian on \(L^2(\Omega)\) by \(\Delta^D\), recall that the classical Dirichlet-to-Neumann operator for \(\lambda\in\mathbb{R}\backslash\sigma(-\Delta^D)\) is the operator in \(L_2(\partial\Omega)\) whose graph is given by \[ \begin{multlined} D_\lambda=\{(g,h)\in L_2(\partial\Omega)\times L_2(\partial\Omega):\\ \text{ there exists }u\in H^1(\Omega)\text{ with }\Delta u+\lambda u=0, u_{|\partial\Omega}=g\text{ and }\partial_\nu u=h\}, \end{multlined} \] where \(u_{|\partial\Omega}\) denotes the trace of \(u\) on \(\partial\Omega\) and \(\partial_\nu u\) is the outer normal derivative. If \(\lambda\in\sigma(-\Delta^D)\), then \(D_\lambda\) is still a subspace of \(L_2(\partial\Omega)\times L_2(\partial\Omega)\) but not the graph of a linear operator anymore. The abstract framework chosen by the authors is the following. Given two complex Hilbert spaces \(V\) and \(H\), a symmetric continuous sesquilinear form \(a:V\times V\rightarrow\mathbb{C}\) and a continuous, linear operator \(j\in\mathcal{L}(V,H)\), the linear relation (i.e., subspace of \(H\times H\)) associated with \((a,j)\) in \(H\times H\) is defined as \[ A=\{(x,y)\in H\times H:\text{ there exists } u\in V\text{ with }j(u)=x\text{ and }a(u,v)=\big(y,j(v)\big)_H\text{ for all }v\in V\}, \] where \((\cdot,\cdot)_H\) denotes the scalar product in \(H\). (By choosing \(V=H^1(\Omega)\), \(H=L_2(\partial\Omega), j:H^1(\Omega)\rightarrow L_2(\partial\Omega)\), \(j(u)=u_{|\partial\Omega}\), the trace operator and \(a(u,v)=\int_\Omega \nabla u\overline{\nabla v}-\lambda u\overline{v}\, dx\), one recovers the above \(D_\lambda\).) The key component of the article is the introduction of the following, seemingly new concept. The sesquilinear form \(a\) is called compactly elliptic if there is a Hilbert space \(\tilde{H}\) and a compact \(\tilde{j}\in\mathcal{L}(V,\tilde{H})\) such that \[ \exists\,\omega\in\mathbb{R}, \mu>0\, \forall u\in V:\operatorname{Re}a(u,u)+\omega\|\tilde{j}(u)\|^2_{\tilde{H}}\geq \mu\|u\|_V^2. \] The authors prove (Theorems 4.5 and 4.15) that the linear relation \(A\) associated with \((a,j)\) is self-adjoint and bounded below if \(a\) is compactly elliptic. Generalising the notion of self-adjointness for operators on a complex Hilbert space, a linear relation \(A\) is called self-adjoint if \((x,y)_H\in\mathbb{R}\) for all \((x,y)\in A\) and if, for all \(s\in\mathbb{R}\backslash\{0\}\), the projection onto the second coordinate of the linear relation \[ A+isI:=\{(x,y+isx):\,(x,y)\in A\} \] equals \(H\). Moreover, a self-adjoint linear relation \(A\) is called bounded below if there is \(\omega\in\mathbb{R}\) such that \((x,y)_H+\omega\|x\|^2_H\geq 0\) for all \((x,y)\in A\). The most important tool in establishing the above result is Lemma 4.1, where it is shown that the compact ellipticity of the (not necessarily symmetric) continuous sesquilinear form \(a\) implies that the operator \(\mathcal{A}:V\rightarrow V^*\) defined by \((\mathcal{A}u,v)_{V^*\times V}=a(u,v)\) is invertible whenever it is injective, where \(V^*\) denotes the anti-linear dual of \(V\). This result can be viewed as a generalization of the Fredholm alternative and is used as a substitute for the Lax-Milgram lemma when the sesquilinear form \(a\) is not coercive. For \(\Omega\) as above, the results of Section 4 are applicable to the symmetric continuous sesquilinear form \[ a(u,v)=\int_\Omega\sum_{k,l=1}^d a_{k,l}\,\partial_k u\,\overline{\partial_l v}+c\, u\overline{v}\,dx \] on \(H^1(\Omega)\), where \(a_{k,l},c\) are real-valued, bounded, and measurable, and the \(a_{k,l}\) satisfy a uniform ellipticity condition, \(j:H^1(\Omega)\rightarrow L_2(\partial\Omega)\), \(j(u)=u_{|\partial\Omega}\), and \(\tilde{j}:H^1(\Omega)\hookrightarrow L_2(\Omega)\). The corresponding linear relation \(A\) associated to \((a,j)\) is then self-adjoint and bounded below, and thus the graph of a self-adjoint bounded below Dirichlet-to-Neumann operator on \(L_2(\partial\Omega)\) if \(0\notin\sigma(\mathcal{A}^D)\), where \(\mathcal{A}^D\) denotes the Dirichlet realization of the operator associated to \(a\). Moreover, the authors investigate strong convergence of the resolvents and \(C_0\)-semigroups associated to a sequence of self-adjoint linear relations obtained in the above manner. Various applications of these results are given and several interesting concrete examples are discussed.