[Ifeffit] amplitude parameter S02 larger than 1
scalvin at sarahlawrence.edu
Mon Mar 23 21:45:04 CDT 2015
The method Ifeffit uses to compute uncertainties in fitted parameters is independent of noise in the data because it, in essence, assumes the fit is statistically good and rescales accordingly. This means that the estimated uncertainties really aren't dependable for fits that are known to be bad (e.g. have a huge R-factor, unrealistic fitted parameters, etc.), but since those fits aren't generally the published ones, that's OK.
Secondly, the high-R amplitude will not be essentially zero with theoretically-generated data, even if you don't add noise, because the effect of having a finite chi(k) range will create some ringing even at high R.
Frankly, the default method by which Ifeffit (and Larch? I haven't used Larch) estimates the noise in the data is pretty iffy, although there's not really a good alternative. The user can override it with a value of their own, but as you know, epsilon is a notoriously squirrelly concept in EXAFS fitting. The really nice thing about the Ifeffit algorithm is that it makes the choice of epsilon irrelevant for the reported uncertainties.
What it is NOT irrelevant for is the chi-square. For this reason, I personally ignore the magnitude of the chi-square reported by Artemis, but pay close attention to differences in chi square (actually, reduced chi square) for different fits on the same data.
Sarah Lawrence College
> On Mar 23, 2015, at 10:18 PM, Anatoly I Frenkel <afrenke2 at yu.edu> wrote:
> Hi Yuanyun and Bruce,
> Your test is a very nice way to check the effect of the muffin tin radius on the fit value of the amplitude fator.
> Although I hadn't run your test myself I would have expected that the error bars produced by the fit of a theory to "theoretically generated data" would have been huge because you haven't added noise to the "data". Or have you? It is not clear why the uncertainties reported by Yuanyun are so small, given that the analysis method uses the high r signal in the data as a measure of noise. Since the data had no noise the high r amplitude should be essentually zero and that would renormalize error bars to huge values.
> I may have misunderstood what was done in this test, and, perhaps, the noise was added automatically as one of the option that Yuanyun chose. Or there was some other reason why errors came out to be small? Thanks for your comments,
> From: ifeffit-bounces at millenia.cars.aps.anl.gov [ifeffit-bounces at millenia.cars.aps.anl.gov] on behalf of huyanyun at physics.utoronto.ca [huyanyun at physics.utoronto.ca]
> Sent: Monday, March 23, 2015 9:46 PM
> To: XAFS Analysis using Ifeffit
> Subject: Re: [Ifeffit] amplitude parameter S02 larger than 1
> Hi Bruce,
> Thanks for the comments.
> Just to make an end to this topic so far: the samples which give large
> amplitude are suspected to be more complicated than the normal
> skutterudites I showned to you last fall. Normal skutterudites data
> actually didn't cause me worried because its S02 is fitted in a
> typical range. In addition, there is one example on the paper PRB 86,
> 174106(2012) which has >3 angstrom nearest distance and the authors
> reports very normal s02 values. Therefore, I will keep tracking down
> why the amplitude is fitted to large and how to relate this amplitude
> to what we found here for my specific case.
> Thank you everyone.
> Anyway, this test result gave me some hints to tracking down why there is
> Quoting Bruce Ravel <bravel at bnl.gov>:
>> On 03/23/2015 04:10 PM, huyanyun at physics.utoronto.ca wrote:
>>> This is to follow up the *test* experiment you suggested. Attached are
>>> three Artemis files.
>>> I chose the Fe (im-3m) structure to do the test. The normal crystal
>>> structure has its first-shell distance at 2.48 angstrom. A large
>>> structure was created so that the first-shell distance reaches 3.11
>>> Both normal and large structure are calculated in JFEFF in the same
>>> manner (i.e., same path number, same calculation procedure, same
>>> sigma2=0.0045 for all paths) to generate the calculated chi data.
>>> File #1 attached was exactly following the procedure you mentioned. The
>>> quick first shell fits very well, four parameters except S02 give normal
>>> results. The amplitude parameter S02 is very large (2.2+/-0.14).
>>> I extended the test in #2 and #3 for comparison. In #2, calculated data
>>> based on large structure is presented to fit to the same large
>>> structure. As expected, we get reasonable fit looking with other
>>> parameters normal. However, amplitude S02 fits to 1.80+/-0.03. In #3,
>>> both data and structure are normal. In this case, it is not surprising
>>> to get good fit and all parameters including S02 turn out to be close to
>>> their true values.
>>> I think it is clear that first-shell distance larger than 3 angstrom
>>> does has effect in making amplitude artificialy large.
>> Very interesting! It's rather difficult to understand how to relate
>> what you have found to your actual sample. That is certainly true in
>> a quantitative sense. I don't think this tells you how to interpret
>> the actual value that you are finding for S02. But I think it does
>> give a hint for why you are getting such a large value in your case.
>> I will be very interested to see how you address this when you publish
>> your results.
>> For everyone else, I would caution against reading too much into what
>> Yanyun has found. (Even more so than I would caution her!) She is
>> looking at a really unusual situation. Her materials really do have
>> extraordinarily long near neighbor distances. In fact, last fall when
>> she showed me her data, I was quite shocked that such a thing exists.
>> In all my years doing XAS and being a beamline scientist, I had never
>> seen such a long near-neighbor distance. The FT of the data was
>> really quite remarkable!
>> Trying to understand a large S02 is a common question on this mailing
>> list and elsewhere. However, I don't think that Yanyun's experience
>> gives much insight to most such problems. Her situation is quite
>> unusual. If you are seeing an oddly large S02, that is something you
>> need to figure out about your sample or your data. In most cases, you
>> cannot explain it away by asserting that Feff made a mistake with the
>> muffin tin radii.
>> As an example, I rather doubt that Yanyun's experience can shed much
>> light on the questions that Jatin Rana was asking over the weekend.
>> Anyway, wow! What an interesting day on the mailing list!
>> Bruce Ravel
More information about the Ifeffit