Hi Zhanfei, Bruce, Scott,

Sorry for jumping in to this conversation a bit late. Like Bruce said, there is no magic trick for determining whether a part of the R-space spectrum comes from one or more scatterers. And while XAFS is sensitive to Z, the sensitivity is weak.

Because it's XAFS School week here in Chicago, I thought I'd go through a test of the Z dependence, and also try out a "trick" (I believe I first heard this from Mali Balasubramanian, but I suspect others may know this trick too) that relies on phase-corrected Fourier transforms, and is somewhat related to Scott's description of Joe Woicik's comments.

The phase-correction "trick": If you correct for the phase-shift, the peak in |chi(R)| should be at the interatomic distance (ignoring subtleties in the XAFS equation). Turning this around, the peak in the phase-corrected |chi(R)| will be at the interatomic distance if and only if the phase-shift applied was correct... which means that Z is correct (to within some uncertainty). How well does this actually work on real data?

To work through these two related ideas, I used ZnSe as a test case -- a very simple structure with a well-isolated first shell, and I have some decent data on it lying around. This also seemed like a useful enough category of analysis, that I thought it would be useful to better document. Scripts and results are at

http://xraypy.github.io/xraylarch/xafs/feffit.html#example-6-testing-exafs-sensitivity-to

(turning this into an Artemis project is left as an exercise for the interested reader). The results of just changing scatterer in the fits are pretty clear, and suggest that Z +/- 2 might be a reasonable rule-of-thumb even when refining, R and S02, at least in this case of a well-isolated first shell. The results might be different for lighter backscatterers, but there are claims, especially in the bio-XAFS literature, that one can distinguish N and O ligands at least in some cases. Still, given that the ZnSe case is so clear, it seems reasonable to stick with the more pessimistic "Z +/- 5" rule-of-thumb, as long as the possibility that one can do better in certain cases (and may do much worse in others!).

The phase-correction approach is interesting in that it asks "is this particular fit self-consistent?" instead of "which of these fits is best?". This independent of the fit quality could make it a useful secondary check of Z and R (much like a bond valence sum can be an independent check on the consistency of N, R, and valence). It does not seem highly accurate on its own -- also suggesting Z +/- 2 or 3 is about as well as one can do without further knowledge of the scatterers. That might be partially related to how well one can actually determine the peak position for chi(R) on a grid of 0.03 Ang, and partly related to the fact that other terms in the EXAFS equation alter the phase. In principle, those could be accounted for -- another exercise for the interested reader.

I don't think the phase-correction "trick" would help Zhanfei -- it will NOT work on a mixed coordination shell. But the approaches described might be useful and/or inspiring to others.

--Matt