[Ifeffit] Re: EXAFS analysis of homonuclear metal clusters

Kelly, Shelly D. SKelly at anl.gov
Mon Aug 2 14:22:21 CDT 2004

Dear mailing list - 

This is a general EXAFS question I decided to post in regards to your
collective thoughts and/or experience with this EXAFS analysis approach.

I am currently attempting to model a number of Fe-S clusters with
nuclearities of 2, 3, 4, 5, and 6. In some cases (where all the Fe atoms
are chemically equivalent i.e. same oxidation state and coordination
geometry) it is easy to approach this by assuming a single Fe absorber
model. When the Fe atoms become inequivalent (oxidation states which may
be a mixture of I, II or III, and  different coordination geometry) I
can average the environments (by including all interactions and ajusting
coordination numbers for the different iron atoms accordingly) again
into a single Fe absorber model. The fits go well, giving realiable bond
lengths (when compared to available crystallographic data), DW's, and
statistics (I am also keeping in mind the number of refined parameters
to ensure a well over-determined fit for each of the models). 

My only hesitation, and main reason I am posting this question, is that
I am not completely sure this is a reasonable assumption as the Fe atoms
are in different oxidation states and will have different E0's. What I
do know is there is a shift of roughly 2 to 3 eV in E0 to higher energy
for every increase, by one, in oxidation state (i.e. Fe(I) to Fe(II)).
Is there any way of determining if this small difference in energy, and
hence E0, between the Fe atoms in the clusters can or cannot be
neglected in modelling the EXAFS? 



Hi Mark,

I liked Bruce's response, but since you asked again.  I will go out on a

I believe that your assumption is justified.  The theoretical models
built using Feff7 or less do not have different Ezero values based on
the valence state of the absorbing atom.    The difference is made by
choosing a slightly different Ezero at the edge, rather than a
difference in K=0 in the theoretical model.  In addition, different
Ezero's become important at low k-values.  Using a kmin of 3.0 inverse
angstrom or greater, I would not expect you to need different Ezero's.
If you move kmin from something like 3.0 inverse angstroms to 1.5
inverse angstroms you might find that your model no longer works with
one Ezero.  Also, Ezero and Delr are correlated.  So say you know delr
then you might be able to "see" a few eV shift in Ezero, assuming that
you are using a lot of low k in the data range.  

If you want to see the shift.  Then I would get my hands on the metal
oxides for each valence state.  And then simultaneously fit each
oxidation state together with the mixture...You might be able to pull
the info out this way.

-As Bruce suggested.   It is unlikely that your data range and fit range
will support more than one Ezero. To verify that only one Ezero is
needed:  Try several models with a each shell having it's own Ezero, or
two Ezeros or one Ezero.  Compare the reduced chi square values of the
different models.  If the reduced chi square value decreases by a factor
of two or more then your data will supports more than one Ezero.  Also
check the uncertainty in the values for Ezero.  If the values overlap
considering the error then these Ezero-values are the same and should be
only one variable instead of two.  

I am sure that you can find EXAFS papers of mixed Fe(II)/Fe(III) oxides
such as magnetite, Fe2O3.  I would guess that they are modeled with a
single Ezero.


More information about the Ifeffit mailing list