Once a fit model has been defined it may be optimised (minimised) using MINIMISE. This calculates numerical derivatives of the least squares surface defined by the fit model, and tries to find a minimum using the maximum descent vector. The NAG routine E04FCF is used to perform the minimisation so NAG documentation may be used for further explanation.
The user may enter an estimate of the total ``distance'' from the initialisation model to the best fit model. (The default values seem to be fine.)
The fitting is weighted by the inverse of the estimated variances of the data points if variances exist and weighted fitting has not been turned off. If variances have not been defined the fitting is un-weighted.
The program outputs the fit value for each iteration. This is a reduced chi-squared value for weighted fitting, and an average squared residual for un-weighted fitting.
The NAG routine is likely to end will one of several ``error'' messages, unless it thinks the maximum has been obtained. Depending on the ``error'' message more iterations may be necessary to find the minimum point.
The program uses the memory to create the fit model at each stage, so any previous memory data is lost. At the end of MINIMISATION it is possible to return to the main menu, use EXCHANGE (the default option), and examine the fitted model.