Reconstructions from boundary measurements on admissible manifolds (Q665016): Difference between revisions

Let \((M, g)\) be a compact oriented manifold with \(C^\infty\) boundary, and let \(\Delta_g\) be the Laplace-Beltrami operator. Then the Dirichlet-to-Neumann (DN) map \(\Lambda_{g,q}: H^{3/2}(\partial M)\to H^{1/2}(\partial M)\) is defined by \(\Lambda_{g,q}f= \partial_\nu u|_{\partial M}\), where \(0\) is not a Dirichlet eigenvalue of \(-\Delta_g+ q\), \((-\Delta_g+ q)u= 0\) and \(u|_{\partial M}= f\). To reconstruct \(q\) from the knowledge of \(\Delta_{g,q}\) is a typical example of \textit{A. P. Calderón's} inverse conductivity problem [Comput. Appl. Math. 25, No. 2--3, 133--138 (2006; Zbl 1182.35230)]. In two dimensions, uniqueness and reconstruction for this problem was proved in [\textit{K. Astala} and \textit{L. Päivärinta}, Ann. Math. (2) 163, No. 1, 265--299 (2006; Zbl 1111.35004)]. In three dimensions, assuming \(M\) is admissible; i.e., embedded in a cylinder \(\mathbb{R}\times M_0\) with metric \(g= c(e\oplus g_0)\), where \(c\) is a smooth positive function and \(e\) is the Euclidean metric on \(\mathbb{R}\), \textit{D. Dos Santos Ferreira} and the authors proved unique determination of \(q\) [Invent. Math. 178, No. 1, 119--171 (2009; Zbl 1181.35327)]. In this paper, the following corresponding reconstruction result is proved. Thoerem 1.2. Let \((M,g)\) be an admissible 3-dimensional manifold. If \(q\) is a smooth function on \(M\), then from the knowledge of \(\Lambda_{g,q}\), one can constructively determine \(q\). Denoting \(\Lambda_{g,0}= \Lambda_g\), the following Theorem 1.1 follows from Theorem 1.2. Theorem 1.1. Let \((M,g)\) be an admissible 3-dimensional manifold. If \(c\) is a smooth positive function on \(M\), then from the knowledge of \(\Lambda_{cg}\), one can constructively determine \(c\). The proof follows the reconstruction procedure of \textit{A. Nachman} in the case where \(M\) is a bounded domain in \(\mathbb{R}^n\) and \(g\) is the Euclidean metric [Ann. Math. (2) 128, No. 3, 531--576 (1988; Zbl 0675.35084)]. Nachman's argument relies in a crucial way on a uniqueness notion for complex geometric optics (CGO) solutions (exponentially growing or Faddeev type solution [\textit{L. D. Faddeev}, Sov. Phys., Dokl. 10 (1965), 1033--1035 (1966); translation from Dokl. Akad. Nauk SSSR 165, 514--517 (1965; Zbl 0147.09404)]). As for CGO solutions, the authors prove that if \[ (-\Delta_g+ q)u= 0,\text{ in }T= \mathbb{R}\times M_0,\;e^{\tau x_1}(u- u_0)\in H^1_{-\infty,0}(T),\tag{1} \] that is, if \(u\) is a CGO solution, then \(f=u|_\Gamma\) satisfies \[ (\text{Id}+\gamma S_\tau(\Lambda_{g,q}- \Lambda_{g,0})) f= u_0\text{ on }\Gamma,\;f\in H^{3/2}(\Gamma).\tag{2} \] Conversely, if \(f\) solves (2), then there is a solution of (1) with \(u|_\Gamma= f\) (\S3. Prop. 2). Here \(\gamma: H^1_0(T)\to H^{1/2}(T)\); \(\gamma u= u|_\Gamma\), \(S_\tau= K_\tau\gamma^*\), \(K_\tau f= e^{-\tau x_1} G_\tau(e^{\tau x_1}f)\) and \(G_\tau\) is the Green operator of \(e^{\tau x_1}(-\Delta_g) e^{-\tau x_1}\). Properties of these operators are explained in \S2. The authors state that this constitutes the main new part of this paper and prove that (2) is Fredholm (\(\gamma S_\tau(\Lambda_{g,q}- \Lambda_{g,0}): H^{3/2}(\Gamma\to H^{3/2}(\Gamma)\) is compact (\S3. Prop. 3)). If \((-\Delta_g+ g)u= 0\) and \(-\Delta_g v= 0\), then the basic integral identity \[ \int_{\partial M}(\Lambda_{g,q}- \Lambda_{g,0})(u|_{\partial M})\,vdS= \int_M quvdV, \] is valid [Dos Santos Ferreira et al., loc. cit.]. If \(u\), \(v\) are CGO solutions, then this left-hand side is known. Because \(u|_{\partial M}\) is obtained from \(\Lambda_{g,q}\) as the unique solution of (2), and \(v|_{\partial M}\) is explicitly given. Taking the \(\tau\to\infty\) limit, \[ \int^\infty_0 e^{-2\lambda r} f_\lambda(r,\theta)\,dr,\quad f_\lambda(\theta)= \int^\infty_{-\infty} e^{2i\lambda x_1} q(x_1, r,\theta)\,dx_1, \] is recovered. Here \((r,\theta)\) are polar normal coordinates in \(M_0\) with center on \(\partial M_0\). Since the curves \(\gamma: r\to (r,\theta)\) are unit speed geodesics in \((M_0,g_0)\), \(\int_\gamma e^{-2\lambda r}f_\lambda(\gamma)\,dr\) is recovered for any maximal geodesic \(\gamma\) going from \(\partial M_0\) into \(M_0\), and for any \(\lambda\neq 0\). This is the attenuated geodesic ray transform of \(f_\lambda\) with constant attenuation \(-2\lambda\) (cf. [Dos Santos Ferreira et al., loc. cit.]). The authors remark that these discussions are possible for any admissible manifold of \(\dim M\geq 3\). If \(\dim M=3\), attenuated ray transform is shown to be invertible [\textit{M. Salo} and \textit{G. Uhlmann}, J. Differ. Geom. 88, No. 1, 161--187 (2011; Zbl 1238.53058)]. Hence we have Th. 1.2 applying results on Fourier transforms (\S4). Let \(g= c\tilde g\) and \(\tilde q =c(q-q_c) \), where \(q=-c^{(n+2)/4}\Delta_{c\tilde g}(c^{-(n-2)/4)})\), then we have \[ \Lambda_{\tilde g,\tilde q}f= c^{{n\over 4}}\Lambda_{c\tilde g,0}\Biggl(c^{-{n-2\over 4}} f\Biggr)+ {n-2\over 4} c^{-1}(\partial_{\nu_{\tilde g}} c)f. \] Theorem 1.1 follows from this equality and Th. 1.2 (\S4).