On Wednesday 07 July 2010 05:55:13 pm Chris Patridge wrote:

If a background is difficult to remove due to short paths lengths, can a background fitting be used in the Artemis fit? I have a large data range giving near two times the ind points even when including the background function. After fitting the data, it gave a number of correlations some of which are negative. The parameters give reasonable values and uncertainties and I feel confident about the model. I know the fit is not great but can I use this as the starting point or is the background just giving me a large fudge factor?

Hi Chris, Not only can it be fit in Artemis -- this is the situation for which that feature is implemented in Ifeffit and Artemis. From a numerical perspective, the problem is that the background and the first part of the data have very similar, overlapping Fourier components. Athena (that is, the Autobk algorithm) uses a very simple approach to trying to distinguish background from data in Foruier space. This algorithm works great for a metal -- the first neighbor is really far away and the Fourier distinction between background and the first coordination shell is unambiguous. You are being very persistant (kudos, by the way!) with one of nature's hardest challenges for the Autobk algorithm. That V-O bond at 1.6 A is just crazy hard for Autobk to deal with. Artemis can, in priciple, do a better job. At the level of Artemis, much more is known about the nature of the analysis problem. At this point we have an actual Feff calculation to tell us what the Fourier components from that crazy-short bond actually are. So, we should be in a good position to make another pass as fitting a spline to the background portion of the data. In general, the approach is to make Rbkg quite small in Athena and lett the extracted chi(k) be pretty suspect. Then use Artemis to re-refine the background spline while also fitting the first (and possibly higher) shells. In you case, I think you are on the right track. As you say, the fit is not perfect, but it also doesn't suck. Except for a little bit around 1.3, it's actually pretty close. As for you correlations, I don't find correlations about 0.3 to be particuylarly troubling. That's not neglible, but look how big the correlation between enot and delr is! (BTW, that the correlations are negative just tells you something about how the second parameter responds when you change the first. Positive correlation means that if you increase parameter 1 a bit, you can increase parameter 2 to compensate. Negative means that you have to decrease parameter 2 to compensate.) So, are you at a good starting point? I vote yes. B -- Bruce Ravel ------------------------------------ bravel@bnl.gov National Institute of Standards and Technology Synchrotron Methods Group at NSLS --- Beamlines U7A, X24A, X23A2 Building 535A Upton NY, 11973 My homepage: http://xafs.org/BruceRavel EXAFS software: http://cars9.uchicago.edu/~ravel/software/exafs/