Dear Ifeffit members, I have some fundamental questions regarding noisy data, and I am wondering how to tell whether the data quality worth doing Fourier transformation/EXAFS fitting or not. For example, the attached MnO chi(k) data becomes noisy from 7 Å-1 when it was measured up to 12 Å-1. The deglitching (left: red -> blue) mitigated the strong dips but the high-k end still has too much noise. I practiced the FT and fitting the 1st and 2nd shells by using the lower part. But My questions here are: 1) How noisy would be too noisy? Like the data between 8 and 10 Å-1 in the attached file, can they still be included for the FT? 2) We can choose the high-k end based on the signal-to-noise ratio, but to what extend? With data being noisy from even 5 or 6 Å-1, can they still be used? 3) For fitting the 1st and 2nd shells, I still lack of clear understanding how the high-k portion can influence. If I measure a set of samples, and one of them has much noisy data so a shorter k-range is picked up for background subtraction and FT. In this case, can I still consider the change or evolution in the fitted parameters systematic? Thank you in advance, Best regards, Yang
Hi Yang:
If this data is not due to a poorly prepared sample that can be improved or
mismatched responses in the detectors then you will simply have to live
with it. The data between 8-10 can certainly be included but with a bit
more car in background subtraction. It looks like there might be a long
wavelength oscillation in k-space. I would also use a window function with
a dk of at least 2 on these data. The noise level is not unreasonable in
the range below 10 A-1 but I would like to see the FTs. As for your
question about how the range of k-space included affects the FT, the wider
the range of k-space included, the sharped the features in the FT. The
information is still there but you will not have as many variable
parameters available for your fits.
Carlo
On Mon, Aug 30, 2021 at 3:33 PM Hu, Yang (HIU)
Dear Ifeffit members,
I have some fundamental questions regarding noisy data, and I am wondering how to tell whether the data quality worth doing Fourier transformation/EXAFS fitting or not.
For example, the attached MnO chi(k) data becomes noisy from 7 Å-1 when it was measured up to 12 Å-1. The deglitching (left: red -> blue) mitigated the strong “dips” but the high-k end still has too much noise. I practiced the FT and fitting the 1st and 2nd shells by using the lower part. But My questions here are:
1) How noisy would be “too noisy”? Like the data between 8 and 10 Å-1 in the attached file, can they still be included for the FT?
2) We can choose the high-k end based on the signal-to-noise ratio, but to what extend? With data being noisy from even 5 or 6 Å-1, can they still be used?
3) For fitting the 1st and 2nd shells, I still lack of clear understanding how the high-k portion can influence. If I measure a set of samples, and one of them has much noisy data so a shorter k-range is picked up for background subtraction and FT. In this case, can I still consider the change or evolution in the fitted parameters “systematic”?
Thank you in advance,
Best regards,
Yang
_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit
-- Carlo U. Segre (he/him) -- Duchossois Leadership Professor of Physics Professor of Materials Science & Engineering Director, Center for Synchrotron Radiation Research and Instrumentation Illinois Institute of Technology Voice: 312.567.3498 Fax: 312.567.3494 segre@iit.edu http://phys.iit.edu/~segre segre@debian.org
Hi Yang, There are many reasons why you would prefer your first-shell fit to have your fit include high-k data, when feasible. I’ll name one, just to give you the idea: Suppose you have a substance where you have a good estimate of S02 (transferability from a similar substance measured under the same conditions), but you don’t know the first-shell coordination number or sigma^2. Coordination number and sigma^2 are correlated, since they both primarily affect the amplitude of chi(k), but they are not completely correlated, as the coordination number has the same effect across the entire k-range, wherease sigma^2 has a much greater effect at greater k (in the EXAFS equation, it’s weighted by k^2). So extending the fit up to higher k reduces the correlation between coordination number and sigma^2, which can be very important to understanding the structure of your material. Of course, there are other ways to reduce that correlation, such as by measuring a series of spectra at different temperatures. But it does provide a good example of why high-k information is useful when you can get it. Best, Scott Calvin Lehman College of the City University of New York
On Aug 30, 2021, at 4:32 PM, Hu, Yang (HIU)
wrote: 3) For fitting the 1st and 2nd shells, I still lack of clear understanding how the high-k portion can influence. If I measure a set of samples, and one of them has much noisy data so a shorter k-range is picked up for background subtraction and FT. In this case, can I still consider the change or evolution in the fitted parameters “systematic”?
participants (3)
-
Carlo Segre
-
Hu, Yang (HIU)
-
Scott Calvin