A stable recovery method for the Robin inverse problem (Q1877745)

Let \(\Omega\) be a connected bounded domain of \(\mathbb{R}^2\) with the boundary \(\partial \Omega\). Let \(\gamma\) and \(\Gamma_N\) be two non-empty open subsets of \(\partial \Omega\) such that \(\partial \Omega = \overline{\gamma}\cup\overline{\Gamma}_N.\) Further, suppose that \(M\) is a part of \(\Gamma_N\). The inverse problem considered in the paper is to determine \(\varphi\) on \(\gamma\) such that the solution \(u\) of the Robin problem (NP) \(\Delta u = 0\) in \(\Omega, \partial u/\partial n = \phi\) on \(\Gamma_N, \partial u/\partial n + \varphi u = 0\) on \(\gamma\) also satisfies \(u = f\) on \(M\). Here \(\phi \neq 0\) and \(f\) are given functions. To solve this inverse problem numerically the authors minimize the so called Kohn and Vegelius cost functional \[ J(\varphi) = \int_\Omega| \nabla u^N(\varphi) - \nabla u^D(\varphi,f)| ^2 + \int_\gamma\varphi| u^N(\varphi) - u^D(\varphi,f)| ^2. \] Here \(u^N(\varphi)\) is the solution of the Robin problem (NP) and \(u^D(\varphi,f)\) is the solution of the Robin-Dirichlet problem \(\Delta u^D = 0\) in \(\Omega, u^D = f\) on \(\Gamma_N, \partial u^D/\partial n + \varphi u^D = 0\) on \(\gamma.\) The authors prove that the minimization problem with noisy data of \(f\) has at least one solution. Picking up any of these for a given sequence of data that converge to the actual ones in \(H^{1/2}(M)\), they prove that the so associated sequence of coefficients converge to that of the exact inverse problem in \(L^2(\gamma)\). Thus, this algorithm is self stabilizing and does not need any additional regularization. Some numerical experiments are given to show the robustness of the method.