[Ifeffit] problem about negative

Bruce Ravel bravel at bnl.gov
Mon Nov 10 09:05:58 CST 2014

Hi Zhanfei,

When I present the interpretation of fitting results in one of my
training courses, I spend a quite bit of time on the problem of
interpreting fitting results, including the results that are somehow
surprising of unphysical.

You are right that a negative coordination number is troubling.  So is
a coordination number with an error bar of about the same size as the
best fit value.  Your "NO" parameter is certainly troubling.

Since I do not know anything about the sample you are measuring, how
it was prepared, or how it was measured, I don't know the resolution
to your problem.  However, I want to suggest that you take a broader
view of what the problem might be.

When you come upon a weird value like a coordination number of -1.9,
there are at least 5 broad categories of things that might be going

   1. You may be misusing the software in some obvious way or you may
      have uncovered a previously unknown bug in the software.  This
      seems to be your assumption, but I suspect that it's not the
      problem here.

   2. You may have made a mistaken assumption about the species of the
      scatterer.  The first plot is the result of a couple of quick
      first shell theory calculation made with Artemis for Ni bound to
      oxygen and sulfur, each at the same distance.  Notice how the two
      are almost perfectly out of phase.   Where you to use the wrong
      one, the fit would correct that by using a negative amplitude.
      (I made the calculation for a Ni absorber, but the same thing
      would happen with a Mo absorber.)

   3. You may have run the theory incorrectly.  I occasionally run
      Atoms with the wrong edge by accident.  Attached is the result
      for a first neighbor oxygen scatterer around Mo with the Feff
      calculation done for the K and L3 edges.  These are not quite as
      far out of phase as the first example, but they are pretty far
      out of phase.  With a large enough E0, they would get far enough
      out of phase that a negative amplitude could numerically correct
      the fit.

   4. There are a number of experimental artifact that can yield
      spurious oscillations in your data.  These sorts of things are
      discussed in many places in the literature and on the Tutorials
      page at xafs.org.  Scott gives a nice summary of many of them in
      section 5.9 of his recent book

   5. Some materials with big white lines and short near neighbor
      distances are simply difficult to deal with.  With a big white
      line, the background function needs to have a lot of freedom to
      follow the data properly at low k.  However, giving the
      background spline a lot of freedom gives it Fourier components
      that are associated with the near neighbor scatterer.  Shelly
      discusses this problem somewhat in her tutorials at

I often tell people that weird fitting values are valuable information
becasue they force you to reconsider all of your assumptions and all
of your experiment.  #2 suggests that you should think very hard about
what you expect to find.  #3 suggests that you may have made a silly
-- but easily fixed -- mistake.  #4 suggests that there was a problem
with the sample or with the beamline and you may need to try the
measurement again, this time being more careful about everything in the
experiment.  #5 is the scariest -- some materials simply defy the
standard algorithms used in the software and require extraordinary
care in processing and interpretation.

As I said, I have no way of knowing how to advise you, but I am fairly
confident that some or all of your problem relates to something I've
said in this email.

Good luck,

PS: If #5 turns out to be the culprit, then you should think hard
about how you ask your question.  You provided a plot of chi(R) in
your original question, which is helpful.  But if the problem is
difficulty in removing the background, a plot of mu(E) and the
background function would be useful.  In fact, a small Athena project
file containing the data in question might be even more helpful.

Good questions get good answers.  Vague or incomplete questions get
vague or incomplete answers.

On 11/09/2014 08:53 AM, ZHAN Fei wrote:
> Dear all:
>   When I get a wrong fit about Mo-O and Mo-S fit of first two shells.
>    ssO                =   0.00729812    # +/-   0.00961052     [0.00300]
>    delrO              =  -0.02968655    # +/-   0.03378397     [-0.05765]
>    enot               =  -5.30590978    # +/-   7.08310376     [-16.71557]
>    NO                 =  -1.88450657    # +/-   1.59438947     [0.5]
>    NS                 =   1.71281926    # +/-   1.08552732     [1.00000]
>    delrS              =   0.00418341    # +/-   0.04904499     [-0.05647]
>    ssS                =   0.00329415    # +/-   0.00688491     [0.00300]
>   name         N       S02     sigma^2   e0     delr     Reff     R
> ==========================================================================
>   S1.1       1.000   1.456   0.00329  -5.306  0.00418  2.41350  2.41768
>   O7.1       1.000  -1.602   0.00730  -5.306 -0.02969  1.54660  1.51691
>    name         ei       third     fourth
> ==========================================
>   S1.1      0.00000   0.00000   0.00000
>   O7.1      0.00000   0.00000   0.00000
> ssO with a negative coordiantion number NO.I searched the mailist there
> is only information about negative ss
> http://www.mail-archive.com/ifeffit%40millenia.cars.aps.anl.gov/msg02326.html(negative
> ss to compensate small amp).but here I even get a negative amp.What's
> wrong can cause a negative amp?
> Thanks for help.
> Sincerely,
>     -----原始邮件-----
>     *发件人:* "ZHAN Fei" <zhanfei at ihep.ac.cn>
>     *发送时间:* 2014年11月9日 星期日
>     *收件人:* ifeffit at millenia.cars.aps.anl.gov
>     *抄送:*
>     *主题:* [Ifeffit] too small bond length in R space
>     Dear all:
>        I encounter a problem of exafs fit in an example of Mo which have
>     Mo-O Mo-S Mo-Mo bond.The problem is that the Mo-O peak in R space
>     has too small bond length (less than 1.4A(+0.5A for phase
>     correction) ).Because when 1.2>rbkg>0.68 there is little change in
>     spectrum of R and k space,and when I increase the value of rmin of
>     forward Fourier transform the peak changes little,so I think it's
>     not a noise.But when I try to fit it use the 1.35A path of Mo-O,I
>     can't fit it well.
>     Could you give me some advise about fit too small bond length,or
>     recommend papers of similar situation?
>     picture of R space is attached.
>     Thanks
>     Sincerely,
>     zhanfei
>     ihep of China
>     R space of spectrum
> _______________________________________________
> Ifeffit mailing list
> Ifeffit at millenia.cars.aps.anl.gov
> http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit

  Bruce Ravel  ------------------------------------ bravel at bnl.gov

  National Institute of Standards and Technology
  Synchrotron Science Group at NSLS-II
  Building 535A
  Upton NY, 11973

  Homepage:    http://bruceravel.github.io/home/
  Software:    https://github.com/bruceravel
  Demeter:     http://bruceravel.github.io/demeter/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: o-s.png
Type: image/png
Size: 43515 bytes
Desc: not available
URL: <http://millenia.cars.aps.anl.gov/pipermail/ifeffit/attachments/20141110/d63eb5cc/attachment.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: k-l3.png
Type: image/png
Size: 44454 bytes
Desc: not available
URL: <http://millenia.cars.aps.anl.gov/pipermail/ifeffit/attachments/20141110/d63eb5cc/attachment-0001.png>

More information about the Ifeffit mailing list