# [Ifeffit] k range

Kelly, Shelly D. SKelly at anl.gov
Wed Apr 28 10:46:16 CDT 2004

```Hi Steven,

In general you can use as small of a k-range as you want but you loose
the number of variable that you can determine and your uncertainty in
those variable gets really big too.    If those trade offs are
acceptable then there is no problem.  I assume that you want to
determine a few variables so Scotts equation is all you need to
determine the minimum k-range.  Artemis also lists the number of
independent points in the log file.  That number has to be greater than
the number of variables you can determine in the fit.  In practice, you
need about twice as many independent points as variables to get
reasonable uncertainties.

But if what you say about your data is true then the metal peak is not
real.  In the chi(k) data the signal for each peak starts at k=0 and
goes until the sigma2 term damps it out, usually around 8 to 10 inverse
angstroms for oxygen, but usually much larger values for atoms with more
electrons.  It is physically impossible for the signal of the metal peak
to only exist between 6 and 8 inverse angstroms.   I would compare the
data set that you are having troubles with to the other data sets that
you indicated gave a consistent peak in the FT.  I would bet that the
troubled data set has some artifact in it that is distorting the FT.

Shelly

-----Original Message-----
From: Steven Sangyun Lim [mailto:steven.lim at yale.edu]
Sent: Tuesday, April 27, 2004 6:31 PM
To: ifeffit at millenia.cars.aps.anl.gov
Subject: [Ifeffit] k range

Thank you all very much for the immediate help.
One more question, may I???
What can the minimum kmax be?
I think it can be about 8 according to Scott.
If I use kmin=2, can it be 7 or 6? The reason why I ask this is
that when I plot R space with my sample, which has been reduced and
needs to be confirmed if there is any metal cluster, the metal peak
shows at kmax=8, but it disappears at kmax=6 or 7. At kmax=10, it is
hard to tell because the spectrum is very noisy (it shows split peak).
When there are extremely small metal clusters (even at CN=0.6), I can
see a distinct metal peak that does not change no matter what k range
and weight I use. I wonder any further advice can be suggested.