[Ifeffit] Noisy spectrum and AUTOBK

[Kelly, Shelly D.]

Hi Gabi,

I need some more information to address your concerns. See below:

> I have a question regarding noisy EXAFS, since I
> analyze biological zinc samples. I have seen Bruce's
> data and it is quiet as a sleeping snail all the way
> till chi(k)=16, however I have a completely different
> situation since I encounter noise at high chi(k) and I
> need an advice from whoever can supply it.
> As I use AUTOBK with emax set to maximum I observe
> chi(k) which is quiet at 0-9, and at 9-12 I obsrerve
> clearly the oscillations but with the addition of
> small spikes resulting from noise. At chi (R ) I see
> the first shell peak at 1.5A with amplitude of 1.
> Adjacent to this peak I observe a small peak at 1A
> with 0.5 amplitude. 

What is emax?  Do you mean that you set Ezero to the maximum of the
first derivative of the xmu data?  What does the background function
look like?  Does it have a large spike right at the beginning?  If so
adjusting Ezero can help reduce that.  Usually kmin should be small like
zero.  The exception being if you have a large white line (peak at the
top of the absorption step).  Then you want to increase kmin to help the
bkg go over the white line

>As I change Rbkg and Knin I cannot
> get rid of the small peak; the left region of the
> small peak is situated between 0.5 to 1A at chi(R ),

If you increase Rbkg to 2.5 you must be able to get rid of the low r
peak.  Although I don't recommend that you do that, for real.  Try it,
just for fun and see.  Often the low r peaks really don't matter.  To
get rid of them, the most robust method is to fit the first shell peak
with a theory then pull the fit back into Athena and remove the
background with the fit as a "standard".    --see below for a recipe.


> and my attempts to keep this region "clean" failed.
> However, if I lower emax to 400, corresponding to
> approx. 10 at chi(K), and Fourier-transform the region
> between 2 to 9 (the quiet region), the small peak
> disappears and the region at Chi (R ) between 0 to 1A
> is "clean", but since I utilize less information the
> peaks at chi (R ) have lower resolution. 
> My questions are:
> 1.	What is the source of the small peak ?
> 2.	Is it a real contribution from scattering ligands ?
> 3.	How should I treat the small peak.
> 4.	Should I get rid of it by all means ? if so, how ?

I don't understand the emax variable.  But, Since the low r peak
disappears as you move the k-range up a bit then I believe the low r
peak is an incorrect bkg removal that can be fixed by including a
standard in Athena.  

Here is a recipe to get a nice background function.  I'm sure Bruce and
Matt have this written down somewhere too.

1)  Fit the first shell in Artemis.  Using a limited k-range where you
are really confident that the oscillations are not distorted by the bkg.
k=[2:9] determine a value for deltaE, deltaR and N or SO2. DetlaR, N or
SO2, should be reasonable.  The value for DeltaE depends on your choice
in Athena.  

2) You have to decide if you original choice of Ezero (in Athena) was
close enough.  If deltaE (from the Artemis fit) is more than +/- 10eV
you need to go back into Athena move Ezero and redo step 1.  The rule
is:  deltaE + Ezero should be somewhere on the edge of the xmu data.

2b) If you can not get a reasonable deltaE then rethink your theory.
Maybe R in the theory is way wrong?  Maybe two shells are needed instead
of one?  Or maybe Ezero in Athena is way off?  Maybe your k-range needs
to be even smaller.  

2c) Still having trouble with step 1? --you can also change the k-weight
for the background removal.  For this first step it is often useful to
crank it up, so that the higher k-region has a nice bkg and then just
don't use the low k-region in the fit to get the theory to "line up"
with the data.  For example: use k=[4:12] in Artemis, with k-weight=3 in
Athena for the background.

3) Now you have a theory for the first shell peak with an Ezero
somewhere on the edge, we want to match-up the data and theory before
the first shell peak.  Go back into Athena, read in the fit in k-space.
Then highlight your data, pull down the standard menu and select the
theory data file.  Redo the bkg by plotting both the theory and the data
in k-space.  They should look similar at low k.  Then plot the FT (using
the same parameters as you did in Artemis) and they should look similar
below the edge.   

3b) If not:  --If they do not cross zero in k-space (at low k-values) at
the same places you need to adjust Ezero.  --If they do not have the
same amplitude you may need to adjust Rbkg slightly.  Remember Rbkg can
not be greater than Kmin in for the data region used in Artemis, but
they can be equal and often are equal.

4) Redo the fit and add some more shells as needed, increase the data
range in k-space.  See how much data you can use and still get
reasonable fit and uncertainties.

5) At the very end you might want to repeat this procedure to make a
pretty picture for your pub.


HTH
Shelly