There are a few points in your description that puzzle me. Have you collected not one but a series of spectra where some external condition (e.g., temperature, concentration, time, pH etc) were varied? If not, PCA cannot be used if only one spectrum was collected containing a mixture of two species. Such spectrum would have only one component, of course - itself. Next, assuming you did use PCA for a series of spectra, without having XANES and/or EXAFS data of test compounds how were you able to deconvolute abstract components that PCA generates into the two species that have meening of XANES or EXAFS data? Without test compounds such deconvolution is not possible unless you used not PCA but a linear combination fit of some sort... And, finally, if you used test compounds that could be reliably reproduced by your 2 principal components, why do you need to do anything else? They are your two species.
Please clarify.
Anatoly
----- Original Message -----
From: ifeffit-bounces@millenia.cars.aps.anl.gov
To: ifeffit@millenia.cars.aps.anl.gov
Sent: Tue Apr 29 03:06:28 2008
Subject: [Ifeffit] Using the amplitude reduction factor as a linearcombination fitting parameter
I have recently collected EXAFS spectra of uranium on a FeS2 surface. Using principal component analysis of the XANES and k3-weighted EXAFS spectra, I have found that there are two uranium species which compose the spectra. As a first tentative guess, I believe these two uranium species are uraninite (UO2(c)) and a uranyl species. I would like now to fit the fourier transform functions (real parts and magnitudes) using the theoretical paths and path degeneracies created by feff, and use the amplitude reduction factor S02 as a fitting parameter to derive the relative amounts of the two uranium species in my samples.
Normally, this S02 is taken as a constant (between 0.7 and 1.0), and the path degeneracies are fitted. So normally, S02 is not really a fitting parameter (some papers derive it even with theoretical functions). However, given the fact that S02 and N are completely correlated, I think it is justified to use this approach.
Can someone comment on this?
Many thanks in advance,
Christophe
SCK·CEN Disclaimer
http://www.sckcen.be/emaildisclaimer.html
