[Ifeffit] Fitting XANES spectrun with standards

[Bruce Ravel]

Adrien,

For starters, I am not very happy with your data.  With such short
pre- and post-edge ranges, I really doubt that you can reliably
normalize any of your data.  Without relilable normalization, I would
doubt any result you get from LCF.  In fact, if presented with data of
such limited range in a manuscript for review, I would be hard-pressed
to recommend publication.

But that is not the point of your question.

You have not given quite enough information about what you did, but I
have a guess as to the "problem".

"Fit all combinations" does a sequence of fits using all combinations
up to some number.  The default is four.  Using this default and your
set of 8 standards, that means the number of fits performed is
8-choose-4 + 8-choose-3 + 8-choose-2 = 70 + 56 + 28 = 154.  That is,
that bit of automation will attempt to find the best fit of the 154
ways of combining 4, 3, or 2 of the standards.  There are various ways
to tweak the total number of fits done in the combinatorial sequence.

As I said, you have not been clear about what steps you took, but it
would seem that you are comparing the results of that combinatorial
sequence of up to 4 standards with the result of fitting all 8
standards to your data.

Those are not and should not be the same.

The combinatorial fitting is just a wrapper around the normal fit.  In
this case, the combinatorial thing just does the normal fit 154 times.

If my assessment is not correct, you did something different, and you
think you have found a problem -- that's fine.  The software gets
better thanks to good bug reports.  Please remember that I need to be
able to reproduce on my computer EXACTLY what you did to expose the
problem. 

B


On Friday, June 29, 2012 10:30:57 AM adrien couet wrote:
> Hi all,
> 
> I am having some troubles when I fit XANES spectra with a library of
> standards. The unknown spectrum has been processed (background removal,
> normalization,...).
> 
> If I understand correctly the documentation, I have two ways to fit the
> unknown spectra:
> 
> - using "fit this group" and I will get the best fit using my standards
> library
> - using "fit all combinations" and I will get all the different fit results
> from the best to the worst.
> 
> However, it appears that the result given by "fit this group" does not
> correspond to the best fit given by "fit all combinations" (it often
> corresponds to a higher R factor fit), and thus "fit this group" does not
> give me the best fit possible. Are these two fitting processes using a
> different fitting algorithm?
> 
> I attached one example to this email. I am using a library of 8 different
> Fe standards to fit the unknown spectra.
> 
> Thanks for your attention


-- 

 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

 Homepage:    http://xafs.org/BruceRavel
 Software:    https://github.com/bruceravel