Kicaj and Matt,
Thanks for the replies. It is certainly reasonable to ask about the
experimental setup, that seems like a good first place to look for oddities.
Sorry about using E0 in two different contexts (Athena and Artemis). I'll
switch to using E0 for the chosen parameter in Athena's background
subtraction and Enot for the guess parameter in Artemis.
I aligned the scans in Athena using the reference foil absorption edge. I
ended up shifting the scans in energy by ~1 eV or less based on the
reference foil. I then set the E0 parameter in Athena to the peak of the
first derivative. This value varied by about 0.5 eV (or less) between the
different samples. This is what I considered when I made the statement about
not seeing the oxidation change in the XANES. There is a shift of about 4 eV
compared to the reference foil, which I would expect because I am looking at
oxides.
I then extracted the chi(k) into Artemis, and used the same Feff
calculations/paths for each sample. In the Artemis fits I am seeing Enot
shifts of 1-2 eV, which is a fair bit larger than the error bars. If I chose
E0 at peak of the first derivative the Enot came out to be ~7 eV. Since this
seemed a little on the big side, I also tried setting E0 to be the top of
the white line, in that case Enot came out to be ~1 eV. The Enot shift
between samples was about the same either way. Could an argument be made
that there is a small shift in the overall oxidation state of the films,
possibly contributing to the changes in the conductivity changes in the
films? Does anyone have any references about using Enot shifts in this way?
Scott - The amplitudes of the XANES features are very close. in the
normalized spectrum the difference practically 0, and in the derivative the
amplitudes are ~5% off. There is bit larger of a difference in amplitude
between the samples (in fluorescence) and a pure In2O3 powder (in
transmission) of ~10% in the first derivative. The samples are sputtered
from oxide sources, so the presence of metallic In should be pretty small. I
am interested about your comment on the relative sizes of the first
derivative peaks. Comparing my samples to the foil the oxide peak amplitude
is bigger than the metal peak by 10% or so. Could this be due to the
differences between fluorescence and transmission or not optimizing the
experiment to measure the foil? Or could it be due to something else more
heinous?
I noticed on individual fits that the dR for the first shell came out nearly
the same (2.162 +/- .007 vs. 2.164) on the two samples, while the Enot's had
the 1-2 eV shift between the samples. I tried to fit the two types of
samples simultaneously while constraining the dR's (and a few 2nd shell
parameters) to be equal to each other and letting the Enot's float. The
relative Enot values came out close to what they were before. If I let dR
values fit independantly, there was no real change present. In other words
the multiple data set fit just made my EB's a bit smaller, but the relative
shifts stayed about the same.
Would there be some reasonable way in a multi-data set fit to constrain the
Enot's?
Brandon
On Mon, Jun 6, 2011 at 9:04 AM, Matt Newville
HI Brandon,
If I understand right (and to echo Darius's questions), it seems like you measure samples with a metal reference, and aligned the spectra in Athena so that the references matched. That's a fine way to go. I would ask: how big were the needed energy shifts? Are the experimental mu(E) spectra aligned well at this point? If you're studying metal oxides with different oxygen content, you might very well see oxidation in the XANES. If I understand correctly, you're saying you don't see this.
Then, you extracted the chi(k) from the aligned spectra. How much did E0 vary for the shifted spectra in this background-subtraction step?
Then, you pulled these chi(k) into Artemis, and see different E0 shifts in the fits. This E0 is a little different, in that it is the E0 shift applied to the Feff calculation to match the experimental spectra. If you use different paths or different calculations, such E0 shifts might happen, and wouldn't be highly meaningful.
I know that's not a complete answer, but hopefully that and Darius's questions will help,
--Matt
PS on Q2: 1/ (kR)^2 vs 1 / kR^2
This is probably either a typo, or a different convention based on the definition of f(k) as the electron scattering amplitude. In the Feff world, it should be 1/kR^2, but using 1/(kR)^2 would just map f(k) to k*f(k). _______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit