[Ifeffit] Query regarding error bars
Bruce Ravel
bravel at bnl.gov
Fri Jan 9 09:27:05 CST 2009
On Friday 09 January 2009 09:12:16 am Bindu R. wrote:
> Could any tell me in a simple language about the error bars returned by the
> EXAFS fitting program?
> what do they exactly represent?
> How is it determined?
> How is the number of iterations decided.
> In addition to R-factor what are the other parameters which determines a
> good fit. For a R-factor ~0.001, if the value of chi2~10,000 and reduced
> chi2 ~ 500, can one consider the fit to be good even if one gets a good
> match to the experimental spectra.
I believe that most of these questions are answered at
http://cars9.uchicago.edu/iffwiki/FAQ/FeffitModeling
The number of iterations is determined from the data. That is, when the fit
converges to within some tolerance, the fit stops and the error bars are
evaluated. There is a hard-wired default for that tolerance and long
experience suggests it is reasonable. There is also a hard-wired upper limit
to the number of iterations, but a decent fit never reaches that number.
The error bars are the diagonal elements of the covariance matrix evaluated
during the fit. Because uncertainty is so hard to properly evaluate (as
discussed in the links in the FAQ), the error bars are rescaled under the
assumption that the fit performed was, in fact, a good fit. Thus the error
bars are a reasonable (although probably conservative) estimate of
measurement uncertainty if you believe that the fit is indeed a good fit.
Again, see the FAQ for some hints about how to decide if a a fit is a good
fit.
If you have additional specific questions, let us know so we can answer them
here and update the FAQ as needed.
Regards,
B
--
Bruce Ravel ------------------------------------ bravel at bnl.gov
National Institute of Standards and Technology
Synchrotron Methods Group at NSLS --- Beamlines U7A, X24A, X23A2
Building 535A
Upton NY, 11973
My homepage: http://xafs.org/BruceRavel
EXAFS software: http://cars9.uchicago.edu/~ravel/software/exafs/
More information about the Ifeffit
mailing list