Re: [Ifeffit] Bond length and Coordination number problems from EXAFS

Hi Fuxiang, How much background do you have in EXAFS fitting? Have you taken a course? Do you have access to a good EXAFS primer book (e.g. Grant Bunker's or Scott Calvin's)? /1) Which is better, to refine only the first shell or more shells. / / The results seems to be quite sensitive on the selection of kmin/kmax and Rmin/Rmax, I cannot judge when can get reliable results (base on chi?) / In general, I would take a kmin that is above the edge and away from any white line because background removal in those areas (white line, or on the rising edge) is suspect. I would take a kmax that avoids data that cannot be distinguished from noise. I picked Rmax on the high side to isolate the 1st peak in the FT. I picked Rmin based on the magnitude of the 1st Feff shell and a need to get enough R-range to do a 6 parameter fit. It was a nervous choice since too low an Rmin could bring in unwanted artifacts. What it comes down to is: do the choices allow a sufficient amount of data for use in fitting; and are they defensible in review. What I considered good fits were based in part on looking for where chi^2 is a minimum and in part from inspection of the results - were the results consistent with a basic understanding of how atom-atom distances should manifest. You applied this as well when you initially wrote and noted your fit was giving Pd-Pd distances shorter than Pd-Ni, and you questioned that. As for fitting higher shells, what is the problem you are trying to answer? Is the Pd distribution statistical or showing evidence of clustering? The higher shells are weak and have multiple scattering as well as back-scattering paths. The error in just the first shell Pd content is large, owing to the strong correlation with the mean-square-relative-displacement. I would expect this to also be problematic in higher shells - more so. So, can you build a defensible argument based on a first peak analysis? Is taking more data an option (going to higher k with less noise)? This was just the Pd data. How consistent is the Ni data fit? /2) You used Rmax=2.56, while Rpd is 2.68, not sure if there's any contradict / The sine term in the EXAFS equation is sin(2kR+delta). What is delta? Did you measure reference foils for Ni and Pd? Where does the first peak manifest in the |FT(k^n * chi(k))| - at what R-value? By how much is the peak shifted from the crystallographic distance for the different standards? / / /3) For the Nni and Npd, you change them manually and check the chi values during fitting, is that right?/ Correct. Try this yourself. I suggest you tabulate and plot them so you can see the different subsidiary minima and maxima and how the fit results vary between them. Perhaps use a broad range to get a sense of what a global minimum might be. Please note: The questions I have posed are intended to stimulate thinking and need not warrant a response. Another curious note: If you look at the chi(k)'s for the 1st shell Feff paths, they are almost 180 degrees out of phase at higher k...a suppressing effect that could make it difficult to study higher concentration mixtures. cheers, -R. On 2/21/2017 6:22 AM, Fuxiang Zhang wrote:

Hi, Robert, Thank you very much for the help. Based on the nominal composition, there should have more than one Pd atoms in the 1st shell and both the R values you got are reasonable. I still have some questions about the fitting 1) Which is better, to refine only the first shell or more shells. The results seems to be quite sensitive on the selection of kmin/kmax and Rmin/Rmax, I cannot judge when can get reliable results (base on chi?) 2) You used Rmax=2.56, while Rpd is 2.68, not sure if there's any contradict 3) For the Nni and Npd, you change them manually and check the chi values during fitting, is that right?

Thanks again

Fuxiang

On Tue, Feb 21, 2017 at 3:30 AM, Robert Gordon mailto:ragordon@alumni.sfu.ca> wrote:

Hi Fuxiang,

I tried fitting just the first shell in your data: kmin 3.5 kmax 11.775 Rmin 1.25 Rmax 2.56

I used just Pd-Ni and Pd-Pd single scatter paths with one amplitude, one E, 2 delR's and 2 sigma-squareds for a 6 parameter fit.

I didn't let Nni or Npd vary in a fit, but did change them between fits. I get what seems to be a double minimum yielding similar chi values as I varied Nni+Npd = 12 i.e. Npd in the range 1.6 - 1.8 or Npd 2.0 - 2.2 yielded similar values Another minimum occurs near 2.4 but sig^2-pd is getting larger than seems plausible.

I did not try changing Ranges or windows to see if that would smooth out the oscillations in the fit versus relative N values

amp ~ 0.7 (ok...but I didn't normalise the data or compare with Pd or Ni foils) delEo -7 (which seems large) Rni 2.54 Rpd 2.68 (both a bit small but would need to compare to Ni and Pd foil fits for systematic error in fitting) ssni 0.0050(2) sspd 0.010(3) (error bars from correlation with changing N - strong correlation between Npd and sspd)

This is not consistent with your model, since it has more than 1 Pd in the 1st shell. The R-values (apart from not having checked reference foils) are within reason for what one would expect from sums of atomic radii.

Hope this helps get you pointed in the right direction as you refine your model.

cheers, Robert

On 2/20/2017 10:57 AM, Fuxiang Zhang wrote:

...

Dear All, I met a problem in fitting my sample. The sample is a very simple crystalline Ni (fcc) with minor Pd. I measured both the K-edge of Ni and Pd and fit with a 55 randomly distributed cluster model (from calculations). The model suggested that there are 1 Pd in the first and second-shell, 6-Pd in the 3rd-shell and 2-Pd in the 4th shell. I have attached the FPJ file here, can anyone help me to have a look.

1. In fitting the K-edge of Pd, most paths are single scattering, but I always get a shorter Pd-Pd length than Pd-Ni, even I tried to include all the paths. Not sure if I made any mistakes in the path parameters settings.

2. If including the minor contribution paths, I use the same parameters for the path: @Ni1.1 Ni1.1@ (acute triangle) and the path: @N1.1 Ni1.1@ (obtuse triangle), as well as single scattering path @Ni1.1@. I know there's a problem (though no much effect on the results), but don't know how to set the first two paths' parameters

3. If the result is correct, which may suggest there is no Pd in the 1st-shell, but how to fit the coordination number of Pd is a problem. Since Pd is only 1/12 in the first shell, while the uncertainty of CN from EXAFS fitting is pretty large. Does it make a sense to refine the CN?

FPJ file is attached

Thank you very much

Fuxiang Zhang

_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov mailto:Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe:http://millenia.cars.aps.anl.gov/mailman/options/ifeffit http://millenia.cars.aps.anl.gov/mailman/options/ifeffit

_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov mailto:Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit http://millenia.cars.aps.anl.gov/mailman/options/ifeffit

_______________________________________________ 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