I went through the XSfit/Fit/FitErrorCalc.cxx code and made changes to match those in the v11 error command added over the last year. The changes included the recent fix for when the soft and hard limits differ (clcunc.f 1.27), the fix for numerical accuracy problems near the hard limits (clcunc.f 1.26), setting the initial trial to a maximum of half the hard limit if the default method gives a value outside the hard limit (clcunc.f 1.23), and the randomized choice of a value between xmin and xmax instead of always picking (xmin+xmax)/2 (clcunc.f 1.21).
Questions still to consider: does v12 handle eigenvalue data correctly (clcunc.f 1.24) ? Are new parameters saved correctly if a new minimum is found (clcunc.f 1.25) ? Should the random number generator be initialized (clcunc.f 1.22) ? What happens if curdel is < ylimit(1) (clcunc.f 1.22) ?
UPDATE: Included the clcunc.f 1.22 changes. The others require discussion with Craig.
Keywords: heasoft, xspec
Tuesday, March 01, 2005
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment