Linearized inverse problem for the Dirichlet-to-Neumann map on differential forms (Q1028273)

Let \((M,g)\) be a compact, \(n\)-dimensional Riemannian manifold with boundary \(i:\partial M\subset M\). The Dirichlet-Neuman (DN) map \(\Lambda^k_g:\Omega^k(\partial M)\to \Omega^{n-k-1}(\partial M)\) is defined on differential forms by \(\Lambda^k_g\varphi=i^*(* dw)\) as follows: let \(w\) be a solution of the the boundary value problem \[ \Delta w=0,\qquad i^*w=\varphi,\qquad i^*\delta w=0. \] For any \(\varphi\), the problem above has an unique solution up to the space of harmonic forms \(w\) such that \(i^*w=0\). Consider \(\Lambda_g:\Omega(M)\to\Omega(M)\) as the natural extension by using the decomposition \(\Omega(M)= \bigoplus_{k=0}^n\Omega^k(M)\). For a symmetric second-rank tensor field \(h\) on \(M\), let \(\dot{\Lambda}_{h}= \frac{d\Lambda_{g+th}}{dt}|_{t=0}\) be the Gâteaux derivative of the DN-map in the \(h\) direction. The author investigate how large the subspace of tensor fields \(h\) satisfying \(\dot{\Lambda}_h=0\) is. This space contains the subspace of potential tensor fields \(f\) represented by \(f=d_s v\) for some \(v\in\Omega^1(M)\) satisfying \(v|_{\partial M}=0\), since the DN-map is invariant under isometries fixing the boundary. For a manifold of even dimension \(n\), the DN-map on \((n/2-1)\)-forms is conformally invariant, therefore, whenever \(k=n/2-1\), the spherical tensor fields also belongs to the annihilating space of \(\dot{\Lambda}_h\). Definition. Let \(0\leq k\leq n-2\) and \(k+1\neq n/2\). An \(n\)-dimensional manifold \((M,g)\) is said to be \(\Omega^k\)-rigid if \(\dot{\Lambda}_h=0\) only for potential tensor fields \(h\). The case \(k=n-1\) is not included because an \(n\)-dimensional manifold is never \(\Omega^{n-1}\)-rigid. The main theorem claims that the \(\Omega^k\)-rigidity is equivalent to the density of the range of some bilinear form defined on the space \({\mathcal H}^{k+1}_{\text{ex}}(M)\) of exact harmonic fields. An inverse problem is posed as follows: To what extent is a Riemannian metric \(g\) on a compact manifold determined either by the data \((\partial M,\Lambda_g)\) for all forms or by the data \((\partial M,\Lambda^k_g)\) for \(k\)-forms with some fixed \(k\)? There are few known results for the inverse problem as listed below: {\parindent=7mm \begin{itemize}\item[(i)] Let \(\varphi:(M,g)\to (M,g')\) be an isometry such that \(\varphi|_{\partial M}= \text{Id}\). So, \(\Lambda_{g'}=\Lambda_g\). \item[(ii)] For any two metrics \(g\) and \(g'\) on a \(n\)-dimensional compact manifold, the condition \(\Lambda^{n-1}_{g'}= \Lambda^{n-1}_g\) is satisfied if and only if \(\text{vol}(M,g')= \text{vol}(M,g)\). \item[(iii)] Let \(n= \dim(M)\) be even. If \(g'\) is conformally equivalent to \(g\) (\(g'=\lambda g\), for any positive function \(\lambda\in C^\infty(M)\)), then \(\Lambda^{n/2-1}_{\lambda g}=\Lambda^{n/2-1}_g\). \end{itemize}} In the last two cases, the converse is known to be true when \(n=2\). In the paper, the author mostly studies the linearized version of the inverse problem. Let \(C^\infty(s^2TM)\) be the space of smooth symmetric covariant tensor fields of second rank on \(M\). By taking a one-parameter variation \(g^t=g+th\), where \(h\in C^\infty(s^{2}TM)\), the DN-map Taylor expansion is written as \[ \Lambda_{g^t}=\Lambda_{g}+ t\dot{\Lambda}_{g,h}+o(t). \] Thus, the Gateaux derivative of the map \(g\to\Lambda_g\) at \(g\) and direction \(h\) is given by \[ \dot{\Lambda}^k_{g,h}: \Omega^k(M)\to (\partial M)\to \Omega^{n-k-1}(\partial M), \quad 0\leq k\leq n-1. \] The linearized inverse problem is posed as follows: Given a Riemannian manifold \((M,g)\), to what extent is a tensor field \(h\in C^\infty(s^{2}TM)\) determined either by the operator \(\dot{\Lambda}_h= \bigoplus\dot{\Lambda}^k_{g,h}\) or by the data \(\dot{\Lambda}^k_{g,h}\) for some fixed \(k\)? The following results have been achieved by the author: Proposition 1. For a tensor field \(h\in C^\infty(s^2TM)\) on an \(n\)-dimensional manifold, the condition \(\dot{\Lambda}^{n-1}_{g,h}=0\) is satisfied if and only if \[ \int_M \text{tr}(h)\,dv_g=0,\qquad \text{tr}(h)= \sum_{i,j}g^{ij} h_{ij}. \] Proposition 2. Assume \(n= \dim(M)\) is even. So, \(\dot{\Lambda}^{n/2-1}_{\lambda g,h}=0\) for every function \(\lambda\in C^\infty(M)\). Proposition 3. If \(h\in C^\infty(s^2TM)\) is a potencial tensor field, then \(\dot{\Lambda}^k_{g,h}=0\) for all \(h\in C^\infty(s^{2}TM)\). In order to enunciate the main theorem, consider the spaces of exact harmonic forms \[ {\mathcal H}^k_{\text{ex}}(M)= \{w\in \Omega^k(M)\mid \delta w=0,\;w=d\alpha \text{ for some }\alpha\in\Omega^{k-1}(M)\}, \] and the map \(F:{\mathcal H}^k_{\text{ex}}(M)\otimes {\mathcal H}^k_{\text{ex}}(M)\to C^\infty_k(s^2TM)\). Thus: Main Theorem. Let \(0\leq k\leq n-2\) and \(k+1\neq n/2\). An \(n\)-dimensional manifold \(M\) is \(\Omega^k\)-rigid if the range of the operator \(F:{\mathcal H}^{k+1}_{\text{ex}}(M)\otimes {\mathcal H}^{k+1}_{\text{ex}}(M)\to C^\infty_{k+1}(s^2TM)\) is dense in the space \(C^\infty(s^2TM)\) endowed with the \(L^2\)-topology. Conversely, if there exists a nonzero tensor field in \(C^\infty_{k+1}(s^2TM)\) which is \(L^2\)-orthogonal to the range of operator \(F\), then \(M\) is not \(\Omega^k\)-rigid.