Fitting parametric curves and surfaces by \(l_\infty\) distance regression (Q2583158)

Consider the problem of fitting the parameters of a parametric curve or surface to a given set of data. Whereas this problem is commonly considered with respect to a least squares norm, it is appropriate in certain applications to use a supremum norm. This leads to the \(l_\infty\) distance regression problem that can be formalized as \[ \min_{a,t_1, \ldots,t_m} \max_{1\leq i\leq m} \| x_i - x(a, t_i)\| _\infty. \] The authors present and analyze numerical methods for the solution of this problem based on Gauss-Newton techniques. In particular they discuss how to handle the difficulties that are introduced by the fact that the exact solution is, in general, not unique.