[Ifeffit] S02 selection from reviewer

[Matt Newville]

Hi Peng,

On Sat, Oct 2, 2021 at 2:50 AM Peng Liu <liupeng5182 at gmail.com> wrote:

> Dear IFEFFIT members,
>
> I am sorry to bother you again. I asked about S02 selection for the first
> major revision. I just received the second revision. The reviewer is not
> satisfied with one S02 value for all our samples.
> "
>
> 1. I am still not satisfied with selected SO2 value (it is set to 0.85).
> SO2 is not transferable between different samples and detection methods. It
> is not possible to use a value obtained from different compound using
> transmission measurement mode to completely different other compound
> measured using fluorescence mode. One method to fix SO2 value is to measure
> diluted solution (to avoid self-absorption) of reference material in
> fluorescence mode. Other is to use multiple spectra fitting for all samples
> of interest (e.g. with Sb(V)) measured using fluorescence mode where SO2
> parameter is the same for all samples.
>
> At the same time I am confident that CN values 5.6, 7.1 and 6.9 correspond
> to CN(Sb-O)=6. I suggest reconsidering the SO2 value for measurements in
> fluorescence mode.
> "
>
> We do get the S02 from a similar reference material measured in
> transmission mode, and our samples were all measured in fluorescence mode.
> It is not possible to measure the diluted reference material in
> fluorescence mode in one or two months. If you could give me some
> suggestions, that would be great.
>


It is always challenging to know what to do when a reviewer insists on
being wrong.   Well, I guess it is even more challenging when they turn out
to be right. ;).

Many different sample preparation and measurement effects can suppress the
overall XAFS amplitude.  There are "theoretical/calculational" terms:
     a) the relaxation of passive electrons to the core-hole that gives S02.
     b) the photo-electron mean-free-path.

that are expected to be "atomic only" and so will not vary with measurement
mode or sample-to-sample (mean-free path might be affected by samples
smaller than a few mean-free path lengths, but the evidence for this is
incomplete).  But, the calculations for these (at least from "easily run
XAFS calculations'') terms are imperfect and may need adjusting to
completely match experimental spectra.  For simplicity, we tend to adjust
S02 but not the mean-free-path.

There is also an amplitude term that may vary beamline-to-beamline (or even
between beam runs) but not between samples or measurement mode:
    c) the actual energy resolution of the beamline.

This could be compensated by adjusting either the mean-free-path term or
S02.  Again, we typically just adjust S02.  In my experience, adjusting the
mean-free-path ("Ei" in ifeffit/Artemis/Larch) is not any better than
folding this into S02.   I'll also say that for the very common situation
of "Si(111) monochromator at a beamline intended for XAFS" especially in
the common 5 to 20 keV range, that spectral resolutions tend to be pretty
close to one another.

And there are terms that can reduce the amplitude that can vary from
sample-to-sample, and some of these are different depending on the
measurement mode.
   d) pinhole effects, important for transmission mode.
   e) over-absorption for fluorescence mode.
   f) detector saturation effects for fluorescence mode measured with
pulse-counting, energy-dispersive detectors.
   g) edge step from normalization, where slowly varying backgrounds can be
different for collection mode.

That is, if you have avoided or corrected for "d", "e", "f", and "g", then
S02 will be transferable between samples, at least those measured with the
same beamtime to account for "c".

All that said, I would expect the reviewer may be correct when saying "I am
confident that CN values 5.6, 7.1 and 6.9 correspond to CN(Sb-O)=6.",
especially if that is a main conclusion of the work.   It appears to me
that the reviewer may not be convinced that effects "e" and "f" were
completely avoided or compensated.

That is, claiming that the uncertainty in coordination number is less than
1 (and claiming that CN of 7 and 6 are significantly different) would
require special attention and confidence that the sample-dependent factors
were carefully addressed.   For example, if two transmission measurements
give CN of 7.1 and 6.9 (say with fitting uncertainty of 1) and a sample
measured in fluorescence measurement gives CN of 5.6 +/- 1, I think it
would be fair to suspect that the effects "e", "f", and "g" could be
influencing that difference.  If those effects are large, then you may need
to convince the gentle reader and the not-so-gentle reviewer that
corrections are done appropriately.

Hope that helps,

--Matt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://millenia.cars.aps.anl.gov/pipermail/ifeffit/attachments/20211002/4c654ce4/attachment-0001.htm>