The minimisation method needs to calculate numerical derivatives. To
estimate the descent direction, small variations are tried in each of
the variable parameters. If the fit is far more sensitive to a small
change in one parameter than the others, then it will only really
search for the best fit for that parameter, whilst the others effectively
remain constrained. To avoid this problem, each parameter has a scale
size of variation. Thus, sensitive parameters have a small scale size
and so are only varied by a small amount. When the parameters are input,
default scale sizes are set. These should be suitable for most purposes,
but sometimes better minimisation will be obtained by altering the
default values. This can be seen at the start of the MINIMISATION
process. The first fit (reduced chi-squared) values correspond to
the
variable parameters of the fit model. If one shows much greater
variation than the others, its scale size is too large.
The CHANGE SCALE command allows the fit parameter scale sizes to be examined and values to be changed. This command itself is a sub-menu with the following commands: