Linear least squares problems with data over incomplete grids (Q2465667)

Let us consider the bivariate problem of surface fitting, where data points lie in the vertices of a rectangular gride and the task of computing a suitable function \(f\) which approximate the data values by a least squares approximation. If in some grid points the data values are missing or inaccurate and have to be excluded, then tensor product methods cannot be applied without changes. \textit{P. Dierckx} [Comput. Math. Appl. 10, 283--289 (1984; Zbl 0579.65010)] developed a method to compute data values such that the solution of the tensor product spline approximation problem using these values is the same as the spline approximation problem using only the given values. In the paper this approach is generalized to arbitrary linear least squares problems and a new method is developed for linear least squares problems with linear equality constraints. Algorithms based on these techniques are given and numerical examples to show the effectiveness of the method are presented.