Hi Alan (et al.)
So my questions are naturally:
1) Could not the Rietveld refinement also be extended to the full data range ?
2) If that was done, would there be any fundamental difference between the two methods of fitting ?
(Yes, I know they also refined on a restricted d-spacing interval containing only Mn-O distances, and that there are other advantages to PDF analysis in identifying features that may not be included in the model).
But if you refine the full data with the same model, can there really be any fundamental difference, if in one case you simply do a Fourier transform to real space ? What am I missing here ?
There are answers to this question on a number of different levels. The data are the same in real- and reciprocal-space and it is a matter of preference which one you work in, but to get the full story you have to analyze both the Bragg and diffuse components. We find the PDF function intuitive and easy to work with, especially when considering small deviations from well ordered crystals. A lot of RMC fits to more disordered materials these days are done directly to S(Q).
Stefan Breuner is correct; most Rietveld refinements remove diffuse scattering with an arbitrary background function. In that case, a "value-added" of the PDF refinement is clearly the information that was in the diffuse component. The InGaAs example is a nice example and there are many more in the Chem Comm review (Dr. Hewat, I will send you a copy under separate cover. Anyone else who wants one, pls drop me a line: [EMAIL PROTECTED]).
Even for a perfect crystal there is thermal diffuse scattering. In the PDF this gives rise to an r-dependent PDF peak broadening which contains information about the lattice dynamics. This can be usefully analysed (e.g., see Jeong et al., Phys. Rev. B 67, 104301 (2003)), and also overanalysed (Dimitrov et al., PRB 64, 14303 (2001)). However, the real value of the PDF is in probing aperiodic disorder when it is present.
The thermal factor argument is a bit tricky. In general, better thermal factors will be obtained from data measured over a wider-Q range. This can become difficult in Q-space when significant Bragg-peak overlap makes the background subtraction uncertain. PDF doesn't suffer from this problem because there is no arbitrary background subtraction. However, Rietveld could be carried out on an explicitly corrected S(Q) and this problem would go away. Thus, in principle, until Rietveld is done on fully corrected S(Q) data the PDF should give more reliable U's. However, I would say that at this point, from a practical point of view, this is an overstatement. We need to fine-tune some things like profile functions in the PDF refinements (currently Gaussians...remember Rietveld in the early '70's?) to fully exploit this. We are working on it.
Thanks for bringing it up...I am delighted that PDF refinements are on the radar screen.
S
-- Prof. Simon Billinge Department of Physics and Astronomy 4268 Biomed. Phys. Sciences Building Michigan State University East Lansing, MI 48824 tel: +1-517-355-9200 x2202 fax: +1-517-353-4500 email: [EMAIL PROTECTED] home: http://www.totalscattering.org/
