[Ifeffit] Problems with EXAFS Fitting of Metalloprotein Zinc Samples

[Bruce Ravel]

Hi Carolyn,

I'm not certain I have anything more useful to say about your specific
problem than what Robert offered, however I do want to make a more
generic comment in response to your earlier post.

In that one, you said:

   "My understanding is that one of the benefits of fitting in R-space
    is that there is a local minima, whereas in k-space there are many
    minima."

I would say that you are mistaken in this.  Regardless of fitting
space, the software uses a non-linear, steepest-descent, minimization
algorithm.  Such non-linear methods are /always/ susceptible to
problems finding the correct minimum.

There is absolutely nothing about doing a Fourier transform that
guarantees a resolution to the problem of finding the correct
minimum.  In fact, that's often easy to demonstrate simply by using
ridiculous initial guesses for fitting parameters.  Use an initial
guess of 20 for E0 and it is very likely that the fit will go to the
wrong place, regardless of fitting space.

When engaging in good practice (i.e. /not/ using 20 as the initial
guess for E0!), EXAFS is pretty safe against the false minimum
problem.  Most EXAFS problems involve the sum of a small number of
damped sine waves -- there's limited chance of the manifold of the
fitting space having a lot of structure around the best fit values.
That said, it is certainly possible to come up with interesting,
complex fitting models with weird correlations between the variable
parameters.

There are at least two big benefits of fitting in R space -- indeed,
they are the reason that R is the default in Artemis.

  1. Specifying the FT range in k and the fit range in R defines the
     bandwidth of the signal, which allows Artemis to approximate the
     number of independent points in the data, which is needed for the
     statistical analysis.

  2. When fitting in R space it is completely clear to the user which
     features in the data are meant to be fitted using a specific set
     of paths from Feff.  That is, the paths used in the fitting model
     should provide all the Fourier components in the fitting range.
     When fitting in k-space, it is more challenging to tell by visual
     inspection when the fit is successful at capturing the details in
     the data.

None of that really addresses your specific problem, but I was a bit
worried about leaving you with what I saw as a misconception about how
the fitting works in general.

Cheers,
B


On 11/10/2015 03:13 PM, Carolyn Carr wrote:
> Hi Robert,
>
> Thanks for your reply. We have done this with standard samples and we
> see the same thing. And of course, we can tell qualitatively from the
> XANES that our sample is not 8 coordinate, and not even 6 coordinate.
>
> We have never tried changing the temperature but we can try this the
> next time we go to SSRL.
>
> We actually do fit multiple scattering features as imidazoles which is
> one of the things we consider as a strength of Artemis for our basic
> needs. This is part of why we think this is strange, because putting 8
> imidazoles around a zinc seems kind of silly but it likes it. 5
> imidazoles and 3 sulfurs and I have an R-factor of 0.2% every time.
>
> If there is non standard multiple scattering that occurs in multiple
> unrelated proteins from different organisms then we would have no idea
> what this could be.
>
> Thanks for your comments!
>
> Carolyn Carr
> _______________________________________________
> Ifeffit mailing list
> Ifeffit at millenia.cars.aps.anl.gov
> http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit
> Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit


-- 
  Bruce Ravel  ------------------------------------ bravel at bnl.gov

  National Institute of Standards and Technology
  Synchrotron Science Group at NSLS-II
  Building 535A
  Upton NY, 11973

  Homepage:    http://bruceravel.github.io/home/
  Software:    https://github.com/bruceravel
  Demeter:     http://bruceravel.github.io/demeter/