Fitting parametric curves and surfaces by L∞ distance regression

摘要: For fitting curves or surfaces to observed measured data, a common criterion is orthogonal distance regression. We consider here natural generalization of particular formulation that problem which involves the replacement least squares by Chebyshev norm. example, this may be more appropriate one in context accept/reject decisions for manufactured parts. The resulting has some interesting features: it much structure can exploited, but generally solution not unique. method Gauss-Newton type and show if non-uniqueness resolved way consistent with exploiting linear subproblem, only allow properly defined, permit second order rate convergence. Numerical examples are given illustrate this.