[Ifeffit] Comparing similar peak positions in Fourier transform with fitting results

Bruce Ravel bravel at bnl.gov
Thu Jul 23 15:12:13 CDT 2009

On Wednesday 22 July 2009 04:04:19 am Hashem Stietiya wrote:
> Greetings to all,
> I am working on Zn binding to Fe and Al oxides and I am trying to fit the
> second and third shells. I have three spectra from three samples that
> differ in a particular treatment. Visually speaking, the Fourier transforms
> for all these spectra show second and third shells at 2.6 and 3.2A
> (uncorrected for phase shift), respectively. I fit the second shell with Fe
> and the third shell with Al. From the fitting results I obtained the Zn-Fe
> and Zn-Al radial distances for the second and third shells. However, these
> Zn-Fe and Zn-Al radial distances were different among the samples. For
> example, the Zn-Fe radial distance was 3.18A for sample 1, 3.44A for sample
> 2 and 3.51A for sample 3. My question is: if the peaks in the Fourier
> transforms are at the same position for all samples, why is it that the
> fitting shows that the radial distances are actually different? NOTE: I fit
> the 1st,2nd and 3rd shells(R range:1.0-3.2A) together.
>  Thank you,
> Hashem Stietiya
> School of Plant, Soil and Environmental Sciences
> Louisiana State University


It is hard to say without examining the details of your problem, but
this sounds like a common sort of problem.

There is nothing about an EXAFS spectrum that actually identifies
itself as being due to a particular type of atoms.  All that the
Fourier transform of your EXAFS data tells you is at what frequencies
(or distances, as you wish...) your spectrum has large Fourier
components.  We than propose a structural model based on a physical
understanding of the sample in question, do a Feff calculation on that
structural model, and use the Fourier components of the Feff
calculation to model the Fourier components in the data.  If it works
and the model yields physically sensible numbers, then we say "yay!".

The problem is that the parameters we use to fit the Fourier
components of the Feff caluclation to the Fourier components of the
data are not independent of one another.  There are correlations.  In
a numerical sense, those correlations mean that the fitting algorithm
might be equally happy with two (or more) different sets of parameter
values, so long as the fitting metric is minimized.

What I suspect is happening in your case is that the parameters used
for the Fe scatterer are not very robust parameters in the sense that
different values can be made to give numerically reasonable fits to
the data.  When this happens, you usually see that the Fe parameters
are highly correlated to other parameters in the problem and that
error bars on the Fe parameters are quite large.  Is that the case?
Are the uncertainties in difference large compared to the different
distances you are seeing?  Are the delR parameters for the Fe highly
correlated with otgher parameters -- specifically other parameters
that effect things in the same range of radial distance as the Fe

When you have an instability in the fit, you need to figure out some
way to address it.  It is hard to say in general what that might be,
but it could include making some assumptions about your structure,
constraining two or more parameters to not be floated independently of
one another, or something else.  In any case, you need to draw on some
other source of knowledge -- possibly the results of other EXAFS
fitting problems, the results of other kinds of measurements on the
samples, the results of theoretical calculations, and so on.

Good luck,

PS: Remember that the XAFS conference is next week.  A lot of the
people who might otherwise answer questions will be fairly unavailable
until the following week.


 Bruce Ravel  ------------------------------------ bravel at 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/~ravel/software/exafs/

More information about the Ifeffit mailing list