[Ifeffit] Fit in TiO2

Bruce Ravel bravel at bnl.gov
Wed Oct 5 16:56:28 CDT 2011

On Wednesday, October 05, 2011 10:08:30 am Kleper Rocha wrote:
> I am trying to fit an EXAFS data for TiO2 that is mainly in anatase form
> with a small quantity of rutile, accordingly with XRD. The material was
> produced by chemical route.
> I would to show that Ti-O1, Ti-O2 and Ti-Ti1 suffered a decrease in the
> coordination number. Both in R and q space the fit is not very good for
> Ti-Ti1, like if missing something?
> I think that I should introduce oxigen vacancies to improve the fit, but,
> it is right? If so, how to make it?

Hi Kleper,

There are a number of things I don't understand about your fit (once I
figured out to unzip the attachment then load the unzipped file as an
Artemis project file -- that was an interesting bit of misdirection!)

Looking through your fit history, it seems as though your strategy is
to fix all of the parameters except for one and fit just that.  That
is a *very* dangerous strategy and is one I think that few of us
so-called experts would recommend.  Doing so purposefully ignores
correlations between parameters.  The fitted value of the one floated
parameter is hard to defend because it is so dependent upon the
specific values of the set parameters.

I am pretty skeptical that you can actually measure as many DeltaR
parameters as you seem to be trying to determine.  For example, the
DeltaR parameters of the two parts of your split first shell of oxygen
atoms are likely to be hard to determine independently because those
two oxygen subshells are so close together (only separated by 0.03
Angstroms).  It seems likely to me that your data won't support
measuring those two DeltaR parameters independently.

In fact, I suspect this is what sent you down the scary road of
floating parameters one at a time.  When floating your highly
correlated DeltaR parameters, I imagine that you found them to be
... well ... highly correlated.  because of that, they probably had
very large error bars and may have settled into surprising values.

Another thing that is suprising to me is that you attempted to float
DeltaR independently for each path, but you used the same sigma^2
parameter for every path.  That is obviously incorrect.  There is no
reason to think that a Ti-O sigma^2 would be the same as a Ti-Ti
sigma^2.  Indeed, Ti and the first shell O are bound by a strong bond
and should have a rather low sigma^2.  The Ti and the second shell Ti
are not bound in a chemical sense and should have a much larger

Of course, needing different sigma^2 parameters for different paths
only increases the number of parameters that you have to consider.

My advice is two-fold:

  1. Read some of the papers by the folks who regularly answer
     questions on  this list.  I would recommend, for starters, that
     you seek out some papers by Shelly Kelly or Scott Calvin.  They
     both have written several papers which discuss the issues
     surrounding the parameterization of DeltaR and sigma^2 for
     different paths.

  2. Think carefully about how the different paths contribute spectra
     weight to the fit.  Think about whether it is actually reasonable
     that particular parameters can be independently determined.  As
     an example, I suspect you will find that you will not be able to
     determine DeltaR and sigma^2 for the two first neighbor oxygen
     paths independently.  You will likely have to tie their DeltaR
     and sigma^2 parameters together.  (That is the purpose of a def
     parameter, by the way.)  Similarly, thin hard about how DeltarR
     and sigma^2 for your multiple scattering paths might be related
     to those parameters for the various single scattering paths.
     think if there are ways to constrain the parameters used for MS
     paths to those used for SS paths.

You asked about oxygen vacancies.  I think there are so many other
problems with you fit thus far that it is premature to think about a
subtle effect like that.  Work on what you've got so far.  Think about
what I've written.  And, if you run into trouble, think about what
question you want to ask on this mailing list -- a well thought-out
question is highly likely to result in some excellent responses!

Good luck,


 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/ifeffit/Demeter

More information about the Ifeffit mailing list