Discretization effects in statistical inverse problems (Q1174448)

This paper deals with a statistical inverse problem where the discretization operator plays an important role. A minimax approach over suitable function classes is used and the derived asymptotic results indicate that for a given number of observations, discretization beyond a certain critical rate essentially will not yield extra information, while at the critical rate of discretization the accuracy of estimation is degraded by a constant factor. Examples of relevant statistical inverse problems are discussed in detail and some interesting applications, e.g. the reconstruction of images observed by positron emission tomography, are considered.