# [Ifeffit] How to judge if the coordination numbers are different?

jrkizews at ncsu.edu jrkizews at ncsu.edu
Thu Nov 12 20:40:36 CST 2009

```Matthew,
Thank you so much for your reply. I definitely will try what you
suggested. Could you please explain to me why an increased chi-square can
be used to argue for a true difference.

"To check the robustness of the differences, you could try doing your
co-refinement with just some of the data, for instance the 1-day 300 plus
55-day 1200 with none of the others."
I will try what you suggested above. I just do not understand how to use
to results to test the robustness of the difference. Could you please
explain to me a little more? Shall I try fitting two data sets, instead of
all of them, in the same way, and see if the chi-square changes?
Again, thank you so much for your insightful message.
Fiona

> My intuitive reaction is that these differences are unconvincing, but that
> itself is unconvincing, being based on gut feel.  I
> assume that the numbers in () are error bars?
> Anyway, you can try fitting the whole set with common,
> concentration-dependent CNs (6 variables) or age-dependent CNs (also 6) or
> just one common pair of CNs (2 variables) by
> slaving some CNs to others.  This is easy to do in Artemis.  It can then
> be argued that if the reduced sum-square ("chi-square")
> goes up on applying these constraints, then you've
> got real differences.  However, it may be that the same effect can be
> obtained by fixing the CNs and letting the ss's float in the
> same way as you've done with the CNs.  To check
> the robustness of the differences, you could try doing your co-refinement
> with just some of the data, for instance the 1-day 300
> plus 55-day 1200 with none of the others.
>
> A possible experimental control, if you can get more beamtime, would be to
> age a 1200mmol sample, then dilute it with more peat to a
> concentration of 300mmol, and age some more,
> while at the same time aging a 300mmol sample.  If these two came out
> indistinguishable, then either the concentration effect is due
> to a reversible reaction or is some strange artifact
> of concentration.  Since you're working right at the margins of what the
> technique can do, you have to consider possibilities like
> that.
>     mam
> ----- Original Message -----
> From: <jrkizews at ncsu.edu>
> To: "XAFS Analysis using Ifeffit" <ifeffit at millenia.cars.aps.anl.gov>
> Sent: Wednesday, November 11, 2009 3:01 PM
> Subject: [Ifeffit] How to judge if the coordination numbers are different?
>
>
> Dear XAS community members,
> I am studying the structural change of ferrihydrite as it ages with peat,
> which is one kind of organic matter. Totally I have three systems. In
> these systems, ferrihydrite/peat ratios are 300, 600, 1200 mmol Fe/kg
> peat. Smaller number means more peat in the background with ferrihydrite.
> I have collected EXAFS data for these three systems that have aged for
> different times. The magnitude portion of the Fourier transformed data
> look very much the same within each system, with small variation in the
> amplitude. I included two Fe-Fe paths (R=3.075 and 3.376 Å) to fit the
> second shell. The coordination numbers of these paths seem to increase for
> each system as it ages. Following is some results:
> 300 mmol/kg              1-day          7-day
>                Fe-Fe1    1.6 (0.2)      1.7 (0.2)
>                Fe-Fe2    1.7 (0.4)      1.8 (0.4)
> 600 mmol/kg              1-day          7-day              55-day
>                Fe-Fe1    1.8 (0.3)      1.9 (0.3)         1.9 (0.4)
>                Fe-Fe2    2.0 (0.3)      1.9 (0.3)         2.1 (0.5)
> 1200 mmol/kg             1-day          7-day              55-day
>                Fe-Fe1    1.8 (0.3)      1.9 (0.3)         2.1 (0.4)
>                Fe-Fe2    2.0 (0.3)      2.0 (0.2)         2.0 (0.5)
> I was told that fitting results of coordination number have big error.
> When I fit these data, I put all data sets together. These data share a
> lot of common variables such as coordination number of first shell Fe-O
> path. The bottom line is that the total number of variables are much
> smaller than the independent points, which increases the accuracy of
> fitting.
> The fitting results coordination numbers are different. However, I do not
> know if the difference is real difference, or they are the same within
> error, or statistically they are just the same. Could anybody share your
> knowledge with me?
> Regards.
> Fiona Kizewski
> North Carolina State Univerity
> _______________________________________________
> Ifeffit mailing list
> Ifeffit at millenia.cars.aps.anl.gov
> http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit
>
> _______________________________________________
> Ifeffit mailing list
> Ifeffit at millenia.cars.aps.anl.gov
> http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit
>

```