Fwd: P-EXAFS fitting results and multiple scattering
Hi Dalton,
I am CCing this to you and to the ifeffit mailing list, which is, of
course, the appropriate venue for asking questions about the software
and about XAS fitting in general. Please use the mailing list in the
future.
First off, kudos for measuring EXAFS data of this quality on a
phosphorous surface complex. That is a non-trivial measurement.
All of your problems (O MS paths, uncertainties, fit quality) are all
inter-related and, I think, could be addressed, at least in part, by
considering a somewhat more sophisticated approach to the data
analysis problem.
For those who haven't opened up Dalton's project file, you'll find
that Dalton has used Artemis' "quick first shell theory" tool to
generate a Feff path for the phosphorous-oxygen path. He then rather
abused that tool to try to get it to make MS paths involving the P and
O atoms.
I am not a big fan of trying to overuse the QFS tool beyond the
intention implicit in its name. If you need to model an unknown
species, you are much better off trying to find a more elaborate
starting model.
That is a topic that Shelly wrote about extensively in her papers from
her days as a uranium biogeochemist. Her concept is that, while you
don't necessarily know quite what your sample looks like, you can make
some informed guesses. Her strategy, then, is to pick a crystal
structure that resembles what you expect to find in the real sample
and run Feff on that. Then parameterize those paths from the Feff
calculation that you think represent your real structure. Here is one
example of this http://dx.doi.org/10.1016/S0016-7037(02)00947-X
In that paper, she models, for example, a uranyl-phosphoryl bond using
some kind of uranyl phosphate mineral. She doesn't use every path
from the mineral -- just the ones that describe the monodentate U-O-P
ligand and its associated SS and MS paths.
One big advantage of this approach is that you will have well formed
potentials for all of your scatterers. Abusing the quick first shell
tool from Artemis pretty pretty much guarantees that you will have a
poorly formed potential surface. That is certainly true for the
5-atom cluster you ended up with in your Feff calculation.
I also think that you need to examine your data rather more
critically. Although your data are impressively good for phosphorous
EXAFS, they are still pretty noisy data. Plot your chi(R) out to 10
Angstroms. Do you see how big the peaks are between 6 and 10 Ang?
That indicates the level of noise in your data. The peaks between 1.5
and 3.5 Ang are not much bigger than the level of the noise.
Regardless of how well you pick your starting model in Feff, there
will be a limit to how much you can say about these data beyond the
first coordination shell.
I am not really clear quite what you did to fit your data. What you
sent to me had all the parameters either set or skipped. Given that
it is not clear how much signal between 1.5 and 3.5 is above the level
of the noise, it is almost certainly not reasonable to think that you
have anywhere near the 14 independent points your k- and R-range
settings would suggest.
You say that your colleagues tend to get smaller uncertainties than
you are seeing. I imagine that is because their data are considerably
less noisy that these P edge data.
That might not be what you were hoping to hear. But data are what
data are. Yours are of limited quality and there will be severe
limits on how much information you will be able to extract from them.
B
---------- Forwarded Message ----------
Subject: P-EXAFS fitting results and multiple scattering
Date: Sunday, April 22, 2012, 05:08:14 pm
From: "daltonabdala@gmail.com"
Ok Bruce,
Thank you so much for your help and input. I kinda had the feeling that the
uncertainties were related to the noisy data. I wanted to hear the opinion
of other users more experienced than myself and see their take on that.
I'll use atoms from now on to come up with the paths instead of relying
solely on the QFS theory, as you said.
Thank you very much,
Dalton
On Mon, Apr 23, 2012 at 2:30 PM, Bruce Ravel
Hi Dalton,
I am CCing this to you and to the ifeffit mailing list, which is, of course, the appropriate venue for asking questions about the software and about XAS fitting in general. Please use the mailing list in the future.
First off, kudos for measuring EXAFS data of this quality on a phosphorous surface complex. That is a non-trivial measurement.
All of your problems (O MS paths, uncertainties, fit quality) are all inter-related and, I think, could be addressed, at least in part, by considering a somewhat more sophisticated approach to the data analysis problem.
For those who haven't opened up Dalton's project file, you'll find that Dalton has used Artemis' "quick first shell theory" tool to generate a Feff path for the phosphorous-oxygen path. He then rather abused that tool to try to get it to make MS paths involving the P and O atoms.
I am not a big fan of trying to overuse the QFS tool beyond the intention implicit in its name. If you need to model an unknown species, you are much better off trying to find a more elaborate starting model.
That is a topic that Shelly wrote about extensively in her papers from her days as a uranium biogeochemist. Her concept is that, while you don't necessarily know quite what your sample looks like, you can make some informed guesses. Her strategy, then, is to pick a crystal structure that resembles what you expect to find in the real sample and run Feff on that. Then parameterize those paths from the Feff calculation that you think represent your real structure. Here is one example of this http://dx.doi.org/10.1016/S0016-7037(02)00947-X In that paper, she models, for example, a uranyl-phosphoryl bond using some kind of uranyl phosphate mineral. She doesn't use every path from the mineral -- just the ones that describe the monodentate U-O-P ligand and its associated SS and MS paths.
One big advantage of this approach is that you will have well formed potentials for all of your scatterers. Abusing the quick first shell tool from Artemis pretty pretty much guarantees that you will have a poorly formed potential surface. That is certainly true for the 5-atom cluster you ended up with in your Feff calculation.
I also think that you need to examine your data rather more critically. Although your data are impressively good for phosphorous EXAFS, they are still pretty noisy data. Plot your chi(R) out to 10 Angstroms. Do you see how big the peaks are between 6 and 10 Ang? That indicates the level of noise in your data. The peaks between 1.5 and 3.5 Ang are not much bigger than the level of the noise. Regardless of how well you pick your starting model in Feff, there will be a limit to how much you can say about these data beyond the first coordination shell.
I am not really clear quite what you did to fit your data. What you sent to me had all the parameters either set or skipped. Given that it is not clear how much signal between 1.5 and 3.5 is above the level of the noise, it is almost certainly not reasonable to think that you have anywhere near the 14 independent points your k- and R-range settings would suggest.
You say that your colleagues tend to get smaller uncertainties than you are seeing. I imagine that is because their data are considerably less noisy that these P edge data.
That might not be what you were hoping to hear. But data are what data are. Yours are of limited quality and there will be severe limits on how much information you will be able to extract from them.
B
---------- Forwarded Message ----------
Subject: P-EXAFS fitting results and multiple scattering Date: Sunday, April 22, 2012, 05:08:14 pm From: "daltonabdala@gmail.com"
To: bravel@bnl.gov Dear Bruce,
I'm a Ph D student in Dr. Sparks group and currently doing my research work on P-EXAFS. I'm particularly interested in examining the surface load, pH and time effects on the formation of P surface complexes at the goethite/water interface.
My EXAFS experiments have been carried out at the X15B beamline at BNL. I've employed Athena and Artemis to do data analysis and fitting. I've come across some issues that I would like to share with you and, perhaps, you could give me some advices or come up with a different procedure that I could use in order to fix some of the issues that have raised me some skepticism about my fitting results and hampered me drawing conclusions.
First, I couldn't find a way to get a better fit for Oxygen multiple scattering.
Second, the uncertainties found for all fitting parameters are way too high if I am to compare uncertainties for other systems that my co-workers are used to deal with.
I'm attaching an Artemis project file so you can take a look at it.
Thank you very much,
- Dalton
----------------------------------------- --
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/ifeffit/Demeter
participants (2)
-
Bruce Ravel
-
daltonabdala@gmail.com