Re: breaking isis when fitting

On Sep 18, 2007, at 6:10 PM, John E. Davis wrote:

> Do you have the parameters constrained to the physically appropriate
> min/max values?  As far as I know, isis will not call the underlying
> function if the parameters are outside their min/max limits.

Yes, of course.  But that presumes that the valid range of parameter 
space is an N-dimensional box for N-parameters.  Valid parameter range 
volumes can be, and frequently are, more complicated, especially for 
some of these more sophisticated model functions.  In which case simply 
choosing upper and lower bounds for N model parameters in ISIS isn't 
going to save you, and you have to rely on the internal logic of the 
code to check for odd cases.  There are a number of model functions 
floating about out there, especially things you'd normally find only in 
XSPEC local model builds, where that isn't the case.  Using subplex for 
an XSPEC local model is usually a pretty good test of how clean and 
careful a coder the model builder is.

-Mike