[Ifeffit] fitting questions

[Bruce Ravel]

On Wednesday 07 January 2004 12:42 pm, k-kupiecki at northwestern.edu wrote:
> (1)  I was wondering (this might be a silly question) whether it was
> considered inappropriate to fit over a range larger than your spectra?
> (i.e. k-range of fit set to 2-15 when in actuality the spectra data only
> goes to ~12.5?)

I believe that Ifeffit will do the right thing.  That is, it will
compute such things as the number of independent points using the
actual data range.

Artemis might do the "wrong" thing, in certain situations.  For
instance, when you change the FT and fit ranges, Artemis recomputes
the number of independent points and reports that in the echo area.
That calculation uses the values entered for the ranges.  That is only
a reporting error, though.  The echo area calculation is not used in
any other way.  Artemis should be more clever and check to see that
the data extends at least to the kmax value you enter and, if not, use
the reaal extent of the data in the echo area calculation.

In short, it should not really be a problem if you do what you say.

Thanks for pointing that out.


> (2) I don't quite understand the following the significance of the
> following statement:
>
>             These are not yet computed quite right in all situations...
>             Chi-square for this data set = 9.01156
>             R-factor for this data set   = 0.01651
>
> When previously in the results the Chi-square and R-factor were given as:
>
>             Independent points          =      11.272460938
>             Number of variables         =      11.000000000
>             Chi-square                  =       4.505779601
>             Reduced Chi-square          =      16.537341618
>             R-factor                    =       0.016508545
>             Measurement uncertainty (k) =       0.007252615
>             Measurement uncertainty (R) =       0.091026879
>             Number of data sets         =       1.000000000
>
> I'm confused by the different Chi-square values, and what values of Chi
> square and R-factor for the data set would Artemis be satisfied with?

Well, in my paltry defense, I was honest about not computing it
correctly ;-)

Artemis takes the second list of numbers from ifeffit.  The other two
are computed by Artemis at the end of the fit.  Artemis seems to get
the R-factor right but is computing the chi-square wrong.  Bugs bugs
bugs.

Why does Artemis redo this calculation?  For a single data set fit,
there is no real purpose and it adds nothing to the log file.  For a
multiple data set fit, it is useful to get a sense of how good the fit
is for each individual data set.  The list of values that I said are
taken from ifeffit are the over-all R-factor and chi-square for the
entire fit.  In the case of a multiple data set fit, then, they are
the aggregate values for all data sets combined.  In an effort to
provide as much useful information as possible, Artemis computes the
R-factor and chi-square for each data set individually.  Apparently,
she only does a good job with the R-factor ;-)


> (3)  Also is there anyway to have Artemis deternine the best k-range (or
> perhaps a way to have it step through different values of kmin and kmax)
> ?  Just wishful thinking on my part probably...

The short answer is "no".

I would consider implementing such a thing, but I am not sure what the
good algorithm would be.  (Well, okay -- a good algorithm within the
confines of current ifeffit and artemis technology.  I understand that
there are Bayesian techniques that can determine the extent of the
data.)  Any good suggestions...?

Stepping through values for the krange would not be too hard to
implement, but is not currently part of Artemis.

Have you taken a look-see at the ZnO example Artemis project files
that Scott Calvin contributed in November?  He discusses the issue of
the stability of a fit with respect to making changes of parameters.

HTH,
B


-- 
 Bruce Ravel  ----------------------------------- ravel at phys.washington.edu
 Code 6134, Building 3, Room 222
 Naval Research Laboratory                          phone: (1) 202 767 5947
 Washington DC 20375, USA                             fax: (1) 202 767 1697

 NRL Synchrotron Radiation Consortium (NRL-SRC)
 Beamlines X11a, X11b, X23b
 National Synchrotron Light Source
 Brookhaven National Laboratory, Upton, NY 11973

 My homepage:    http://feff.phys.washington.edu/~ravel 
 EXAFS software: http://feff.phys.washington.edu/~ravel/software/exafs/