Direct numerical identification of boundary values in the Laplace equation (Q1872938)

The authors consider a two-dimensional boundary value problem for the Laplace equation in which, on different parts of the boundary, a varying number of boundary conditions (of all classical types) is given - including the case that somewhere no conditions are prescribed. The aim is to find the Dirichlet and Neumann data on the whole boundary. A quadratic object functional is defined (containing a regularization term) and shown to be strictly convex as a function of the missing Dirichlet data. Using the adjoint problem and the condition that the gradient of the functional be zero on the minimizer, and approximating the given and searched boundary data by finite elements, a numerical method is set up. This leads to a linear system of equations. In a numerical example, solving the inverse problem starting from data obtained from a direct problem, the deviation between identified and original boundary data (i.e.\ without noise) seems acceptable.