Hi Dan, JAQN on the comparison of reduced chi-squares from different beamlines: In order to calculate a chi-square, you need an estimate of the measurement error ("epsilon"). Although you can set that in Ifeffit (and Artemis) to any value you think reasonable, the default behavior of Ifeffit (and thus Artemis) is to make an estimate based on the value of the Fourier transform way up above where there should be any EXAFS signal (as I recall, it uses 15 to 25 angstroms). Although in ideal circumstances this might be reasonable, in practice it can cause difficulties, particularly when comparing data from two different beamlines. As an example, I've worked on beamlines where the lock-in wasn't quite working right, and there was therefore a little oscillation superimposed on the data with a period of a few eV's. This doesn't actually interfere with the EXAFS analysis much (although it could mess up XANES), because it basically just creates spurious signal high in the Fourier transform. But it <italic>does</italic> cause an overestimate of epsilon by Ifeffit, which results in a <italic>smaller</italic> reduced chi-square than appropriate. That's right, taking data on a beamline that is misbehaving can in some cases cause anomalously small chi-square values. (Of course, in the <italic>really bad</italic> scenario that there are spurious oscillations right in the same range as the EXAFS and not above, Ifeffit will give anomalously large chi-square values. But in that case the data's junk anyway.) So what's the bottom line? If you have some clever way of estimating epsilon for each beamline, you can put that into Artemis and compare the reduced chi-squares. But if you don't (and I'd love to hear from people if they do), then I wouldn't use reduced chi-square to compare data from different beamlines; the r-factor will have to do. Actually, in my opinion, reduced chi-square is even a dangerous measure if you are comparing different k-ranges or k-weights, unless you fix expsilon. On the other hand, it's a perfectly good and appropriate measure for comparing different r-ranges, guessed parameters, constraint schemes, etc., etc., because in those cases ifeffit will use the same epsilon for each fit by default. In your particular case, it is a greater concern that some of the values Ifeffit finds for the guessed parameters are significantly different from each other (i.e. they fall outside each other's error bars). --Scott Calvin Sarah Lawrence College At 08:36 AM 10/28/2004 -0700, you wrote:
Also, the data on the same sample from two different beamlines
seems to
be somewhat different, namely the Reduced chi-square is very different
(75.6 in one, 636.4 in the other). In both cases I use the first 3 single
scattering paths. I may have included too much info, but better too much than
too little.