[Ifeffit] k-weight in Artemis

Scott Calvin SCalvin at slc.edu
Tue Sep 26 15:45:02 CDT 2006

Hi Juan,

Excellent questions!

At 03:52 PM 9/26/2006, Juan Antonio Maciá Agulló wrote:

>I read in Artemis manual that a good fit must be consistent for different
>k-weights. When two fits (one with k1 and other with k3) are consistent? I
>mean, which is maximum allowed difference, 1%, 5%...between these two fits?

Ideally, the fitted parameters should have 
uncertainty ranges that overlap. For example, one 
bond length might be 2.03 +/- 0.04 angstroms, and 
the equivalent bond length using a different 
k-weight might be 2.09 +/- 0.03 angstroms.

>I have done different fits changing fit 
>k-weights and k-weights in the plotting
>options. Usually, for k-weights = 1 or 2, I have obtained very strange spectra
>(why?) and then I have decided to change only k-weight in the plotting options
>to 3. After that, spectra looked more like before. I am not sure if this
>gimmick is correct. I mean, can I use different k-weights for fit and for
>plotting? I have obtained "consistent" fits with fit k-weights of 1 and 3, and
>both plotted in k-weight 3, that's ok?

What do you mean, "strange" spectra? They will 
look different for different k-weights, of 
course. That's not a problem. In general, it's OK 
to fit in one k-weight and plot in another, and 
there are a variety of good reasons for doing 
that: for example, you're comparing to published data with a certain k-weight.

>Sometimes when I have changed fit k-weight from 3 to 1, I have observed that
>two high correlations (between ss and S02, and 
>delR and delE0) have dissapeared
>(<0.85). What does it mean? Parameters obtained with fit k-weight = 1 are more

Umm...not more reliable. Just less correlated. :) 
You can look at the uncertainties on the 
parameters to see if that also was reduced when 
you changed k-weights. Sometimes one k-weight 
manages to break correlations more effectively 
than another, but there isn't a simple pattern to 
it. Different parameters are weighted by 
different powers of k in the EXAFS equations, and 
different scattering elements have large 
amplitudes at different values of k, so the 
overall effect is sometimes difficult to predict: 
it's best to just try it and see what happens. 
Ideally, the different k-weights will yield 
similar values for the parameters, although the 
graphs may look quite different.

>And another question, I have seen an option to 
>fit background in Artemis, is it
>obligatory to fit the background to publish EXAFS fits?

No. This is another one of these issues where 
people can reasonably have different preferences. 
Personally, I prefer not to use the Artemis 
background refinement in my published fits, 
partly because it's one more processing step to 
explain and justify. But I do use the ability of 
Artemis to do background refinement to test if my 
Athena background is OK: if fitted parameters 
correlated highly with Artemis background 
parameters, I know there's a problem, and I go 
back and look at the background subtraction 
process in Athena again. If they don't correlate 
highly, then the Artemis background wasn't doing 
anything except improving the visual aesthetics 
of the fit anyway, and I turn it back off.

--Scott Calvin
Sarah Lawrence College

More information about the Ifeffit mailing list