Resolution of regularized output least squares estimation procedures (Q1354174)

The authors study in particular the finite-dimensional problems arising as the finite element approximations to distributed systems. Such systems are closer to those models actually used in numerically implementing the solution of the inverse problem. The considerations of the interior optimal estimators allow to obtain results through the application of estimates of solutions of the optimality system and the implicit function theorem. The formulation is given for the finite-dimensional approximation of the original problem and the conditions implying uniqueness of interior optimal estimators are discussed. A relation between the perturbations of the data and those of the associated optimal estimators is also determined. Using the optimality conditions and the fact that they implicitly define the optimal estimator, the authors study the effect of perturbations of the data on optimal estimators by means of the implicit theorem. For finite-dimensional problems the authors show certain classes of problems in which perturbations of the data have no effect no matter how large. In Section 2 the finite-dimensional problems are posed and the equations for the derivatives with respect to data determined. In Section 3 the recovery function and its differentiability are considered. Finally, Section 4 is devoted to reporting the results of several simple numerical experiments.