[Ifeffit] Linear combination fitting reports

[Bruce Ravel]

On Wednesday 02 September 2009 04:13:55 am Jens Kruse wrote:
> Hi,
>
> I have only two short questions. Is it possible that values for
> chi-square and reduced chi-square given in the "Fit results" and the
> "Combinatorics" page are inverted within Athena?

I tried running a combinatorial LCF series just before writing this
email and I think these numbers are reported correctly.  If you think
otherwise, you will need to show me an example.

Of course, the values for both R-factor and reduced chi-square are
rather peculiar in Athena's LC fitting, as explained in the manual:

  http://cars9.uchicago.edu/~ravel/software/doc/Athena/html/analysis/lcf.html


> In the CSV report  for all fits, which of these both columns ("chisqr",
> "chinu" ) is chi-square and which is reduced chi-square?

I don't think these are reported wrongly.  As I discuss in the manual,
it is hard to know how to evaluate chi-nu in a XANES fit.  I use
epsilon=1 (for want of a better number) and I use the number of points
in the data as the number of independent measurements.  There are
typically of the order of 100 points in a XANES scan, so chi-nu tends
to be about 2 orders of magnitude smaller than chi-square, which
itself is very small due to the value of epsilon.

All of this just underscores the squishiness of the statistical
parameters in this context.  Interpreting LCF results requies the use
of prior knowledge and a very conservative interpretation of what
Athena reports.

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/