Advice on Model Fitting

Finding a optimum fit between a parameterised model and measured data can be a difficult task owing to the problem of local minima, or simply owing to the size of the computational problem being undertaken. To avoid local minima and to reach quickly the global minimum good initialisation is vital. Great care should be taken with the the initialisation process. Many of the commands in the main menu may be useful to obtain good estimates of model parameters. The CONSTRAIN command may be used to set parameter values individually.

The fewer variable parameters, the faster will be the fitting. Thus if certain parameters can be given a good initialisation, it will be useful to constrain these parameters whilst optimum values are found for the remaining parameters.

If the model makes the fit much more sensitive to variation in one parameter than the others, it may completely dominate the minimisation process, such that the other parameters are not optimised. The CHANGE SCALE command allows this problem to be reduced (See Section 17.4, Page ).