Manuscript comments regarding EXAFS modeling
Dear Ifeffit members, I received the following two comments. " Comment 1: Authors have fixed the amplitude reduction factor (SO2) to a fixed value (0.85). This factor is specific to particular chemical compound and sample preparation and quality (mostly homogeneity), measurement method (e.g. absorption, fluorescence). Authors can find in literature [e.g. Rehr2000] that SO2 for ideal samples (having no other effects) represent multielectron effects, which by definition depend on valence and ligands. Even more, SO2 is correlated with Debye-Waller factor (σ²) and coordination number (CN), so any chosen value will be compensated by CN and σ². As coordination numbers are used as quantitative indicators in discussion and following conclusions. I would request to clarify the selection criteria for SO2 values and advise to revise this approach (i.e. not to fix SO2 as the same value for all samples). I do not expect drastic changes in obtained CN values, but this should be tested. Comment 2: As I mentioned previously, coordination number (CN) is correlated with Debye-Waller factor (σ²). My question is: how this correlation is managed (eliminated)? Most probably (in FEFFIT) this is done by using 3 separate values for n (1,2,3), where n is a power in expression chi(k)*(k^n). " I used Artemis for the calculation. 1) Because S02 and CN are multiplication relations in the EXAFS equation, as we usually do, we fixed S02 to obtain CN for unknown samples. 2) there are outputs regarding the correlation between different fitting parameters from Artemis. Is there a way to manage or eliminate the correlation as the reviewer mentioned using Artemis or Larch? If you also could give me some suggestions to answer the comments, that would also be greatly appreciated. -- Best Regards, Peng Liu School of Environmental Studies China University of Geosciences, Wuhan, Hubei Province, PR China https://scholar.google.com/citations?user=qUtyvokAAAAJ&hl=en http://grzy.cug.edu.cn/049121/zh_CN/index.htm
Was there a standard measured that's chemically similar to the unknown? If so, S0^2 could be set by requiring that the CN of the standard be what it's known to be. Often, the dσ² is mostly relevant in terms of differences or changes so an error in the absolute value isn't very important. If valence and ligands make such a big difference then it would be impossible to measure CNs on samples for which no extremely-similar model compounds are available. Also, it's never been clear to me how doing the fit at three different k-powers is better than doing it at just one intermediate value. The idea that it somehow gets you low-k and high-k information in a better way seems like an attempt to pull out more independent data points than actually exist. OTOH, proving that you get the same answers, within error bars, at different exponents would be a confidence-builder. mam On 8/27/2021 7:40 PM, Peng Liu wrote:
Dear Ifeffit members,
I received the following two comments.
" Comment 1: Authors have fixed the amplitude reduction factor (SO2) to a fixed value (0.85). This factor is specific to particular chemical compound and sample preparation and quality (mostly homogeneity), measurement method (e.g. absorption, fluorescence). Authors can find in literature [e.g. Rehr2000] that SO2 for ideal samples (having no other effects) represent multielectron effects, which by definition depend on valence and ligands. Even more, SO2 is correlated with Debye-Waller factor (σ²) and coordination number (CN), so any chosen value will be compensated by CN and σ². As coordination numbers are used as quantitative indicators in discussion and following conclusions. I would request to clarify the selection criteria for SO2 values and advise to revise this approach (i.e. not to fix SO2 as the same value for all samples). I do not expect drastic changes in obtained CN values, but this should be tested.
Comment 2: As I mentioned previously, coordination number (CN) is correlated with Debye-Waller factor (σ²). My question is: how this correlation is managed (eliminated)? Most probably (in FEFFIT) this is done by using 3 separate values for n (1,2,3), where n is a power in expression chi(k)*(k^n). " I used Artemis for the calculation. 1) Because S02 and CN are multiplication relations in the EXAFS equation, as we usually do, we fixed S02 to obtain CN for unknown samples. 2) there are outputs regarding the correlation between different fitting parameters from Artemis. Is there a way to manage or eliminate the correlation as the reviewer mentioned using Artemis or Larch?
If you also could give me some suggestions to answer the comments, that would also be greatly appreciated.
-- Best Regards,
Peng Liu
School of Environmental Studies
China University of Geosciences, Wuhan, Hubei Province, PR China
https://scholar.google.com/citations?user=qUtyvokAAAAJ&hl=en https://scholar.google.com/citations?user=qUtyvokAAAAJ&hl=en
http://grzy.cug.edu.cn/049121/zh_CN/index.htm http://grzy.cug.edu.cn/049121/zh_CN/index.htm
_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit
Hi Peng,
This will echo much of what Matthew Marcus wrote:
For comment 1 on S02, picking a value of 0.85 seems reasonable. I think the
reviewer is asking you to explain how you got that value. Saying something
like "we chose that value so that the data from a simple metal foil (or
simple metal oxide, etc) gave the expected first shell coordination number"
should be enough. Typically, S02 is determined once for a given set of
data with the same beamline conditions and *not* varying from sample to
sample.
For comment 2 on correlation, I would emphasize that N and sigma2 are
well-known to be correlated and that this correlation cannot be
eliminated. The correlation is "managed" as all correlations are managed -
by a statistical analysis of that correlation. The uncertainties reported
for all fitting variables always take those correlations into account.
That is just a normal part of the analysis.
There is a common misconception that using multiple k-weights "eliminates"
correlations between variables. It does not. It is available in
ifeffit/artemis/larch to try to help find more robust solutions. In my
experience, simultaneously using k-weights of 1, 2, and 3 does not
actually givr very results compared to using a k-weight of 2 or 3 alone,
though I'm willing to believe that there are cases where it can help find a
solution for a fit with both low-Z and high-Z scatterers. That is, using
multiple k-weights is a fine thing to do but it does not lower correlations
between N and sigma2 (or E0 and R) by very much.
Cheers,
On Fri, Aug 27, 2021 at 7:41 PM Peng Liu
Dear Ifeffit members,
I received the following two comments.
" Comment 1: Authors have fixed the amplitude reduction factor (SO2) to a fixed value (0.85). This factor is specific to particular chemical compound and sample preparation and quality (mostly homogeneity), measurement method (e.g. absorption, fluorescence). Authors can find in literature [e.g. Rehr2000] that SO2 for ideal samples (having no other effects) represent multielectron effects, which by definition depend on valence and ligands. Even more, SO2 is correlated with Debye-Waller factor (σ²) and coordination number (CN), so any chosen value will be compensated by CN and σ². As coordination numbers are used as quantitative indicators in discussion and following conclusions. I would request to clarify the selection criteria for SO2 values and advise to revise this approach (i.e. not to fix SO2 as the same value for all samples). I do not expect drastic changes in obtained CN values, but this should be tested.
Comment 2: As I mentioned previously, coordination number (CN) is correlated with Debye-Waller factor (σ²). My question is: how this correlation is managed (eliminated)? Most probably (in FEFFIT) this is done by using 3 separate values for n (1,2,3), where n is a power in expression chi(k)*(k^n). " I used Artemis for the calculation. 1) Because S02 and CN are multiplication relations in the EXAFS equation, as we usually do, we fixed S02 to obtain CN for unknown samples. 2) there are outputs regarding the correlation between different fitting parameters from Artemis. Is there a way to manage or eliminate the correlation as the reviewer mentioned using Artemis or Larch?
If you also could give me some suggestions to answer the comments, that would also be greatly appreciated.
-- Best Regards,
Peng Liu
School of Environmental Studies
China University of Geosciences, Wuhan, Hubei Province, PR China
https://scholar.google.com/citations?user=qUtyvokAAAAJ&hl=en http://grzy.cug.edu.cn/049121/zh_CN/index.htm _______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit
participants (3)
-
Matt Newville
-
Matthew Marcus
-
Peng Liu