Automatic curve-fitting-ll. Linear equations

Abstract In this second paper of a three-part series, the automatic FORTRAN IV curve-fitting program, CURFIT, that can handle both linear and non-linear equations is described and demonstrated with linear equations. In CURFIT the maximum tolerance (or uncertainty) for each item is an integral part of the data. An automatic pattern recognition component unambiguously identifies “wrong data-points” and determines whether the data are described by a single function or by overlapping functions of the specified type. The “goodness of fit” criteria are the number of equations with overlapping domains; the median values for all the parameters and their associated maximum errors; and the rejected data-points. CURFIT can be used to unambiguously compare different multi-variable equations, models or theories. It can also be used to detect regions where one of several competing processes predominate. Chemical engineering and chemical data are used to demonstrate the potential of CURFIT.