Hi Juan,
On Fri, Sep 10, 2010 at 7:29 PM, Lezama Pacheco, Juan Salvador
Reposting, just notice no line breaks, sorry! Juan Dear All: First of all, nicely done you guys! This is indeed a set great of tools the community really appreciates it and I have been putting into heavy use. I have and issue though that has been making me scratch my head from time to time. I've been trying to figure out the feature of k-space and R-space fitting "relationship" for Ifeffit.
I usually work out my fits in k-space although I have to admit that R-space fitting in combined approach will definitely be valuable. Now, as I am trying to work some data for Uranium (namely uranyl acetate standard) I am a little bit confused on the real independency of k-space and R-space fitting. This comes from the fact that even if I specified the fitting range in k (fitting in k space), the dk value seems to have an effect in the current fit values (changing as large as 15%).
I'm not certain I understand the questions well enough to answer, but the "relationship" is simply whether a Fourier transform is applied to the data and model.
I understand the issue if the actual fitting range is then (kmin-dk/2,kmax+dk/2) but is this the case? Then what is the real relationship-independency between k-space and R-space fitting? For q-space that is of course another story.
The window function and k-weighting are applied for fits in k-space, so that one may emphasize and de-emphasize different spectral components. Because I'm not sure what you're asking, I'll stall by trying to explain my philosophy for generally preferring to fit in R rather than "unfiltered" k-space and "filtered" k- (aka q-) space. A fit in R-space forces you to actively select the k- and R-range of the fit. A fit in k-space does not limit the R-range, and so is equivalent to using the full R-range of the spectra. Again, one can still apply a k-window and use k-weighting in a k-space fit, of course: it simply doesn't do a Fourier transform of the data (and model). Limiting the R-range of the fit is preferable because the structural model will generally be limited in R (you don't include shell out to 15 Ang in the model), and so you will have data in your for which you are not even trying to supply a model. Remarkably, this (k-space fitting) can work OK in many cases, especially when you are actually modeling most of the spectra, as when there is only one shell that gives signal above the noise. A fit in q-space should be nearly equivalent to a fit in R-space, as you have to select a k- and R-range. Here the preference for R-space is simply that back-transforming is simply not needed. In the distant past, the fitting was necessary to isolate model spectra from experimental standards, and so Q-space fitting was the norm. With Ifeffit and especially Artemis, it is possible to switch back and forth between doing fits in k- and R-space. It's possible (though perhaps not trivial) to fit in both k- and R-space at the same time. I'm not sure that helps answers your question. If not, let us know.
For the record I am using ixafs 1.2 on Mac OS X 10.5.8. Perl distribution seems to crash "unexpectedly" every time I tried the new 3.0.2, but that is another story in another thread...
Yes, sorry and thanks for your patience. I'm trying to build a version that can run on 10.5.* but it's harder than I thought.. --Matt