Dear all,

so we are having a philosophical discussion - after being distracted all
day it is evening, my users seem happy and I have some time to add my 5c
worth to the discussion. For those who do not know me, I am running NPDF
at the Lujan Center and we aim to bring PDF (neutron PDF) to the world !

(1) Rietveld vs. total scattering

Although intentions might have been different and GSAS has the Debye
formula implemented and provides a physical background to describe diffuse
scattering, 99.999% of the cases use some background polynomial and Bragg
intensities and positions are refined to an *average* structural model.
Leaving profile analysis and texture aside, that is the structural
picture. And although the Patterson can be extended to include non integer
hkl's it usually does not and it gives a map of atom atom distances
corresponding to the average structure. PDF Fourier transforms it all so
you get an extended picture of the local-, medium range and long range
atom correlations.

Philosophy is nice - but there are many confused ideas about PDF and
Rietveld and Bragg vs. diffuse scattering, that I think it helps to keep
this distinction clear !

(2) Total scattering

Now we are looking at all of it Bragg and diffuse scattering. Basically
you can analyze in Q or r space - should be the same, except for the
weights as pointed out by someone earlier. Some of the RMC studies refine
data in Q and r simultaneously and more recently add the Bragg intensities
as a constraint as well (using RMCprofile - www.isis.rl.ac.uk/rmc).

So if you have a disordered system (disordered as in deviations from the
average structure) Rietveld along will not give you the answer. Strange
large adp's or their behavior as function of temperature will give you a
clue as to 'there is something going on' but that is it.

(3) Refining S(Q) vs. G(r)

First on synchrotron beamlines or at spallation neutron sources where one
can access high Q (> 40 A**-1), termination ripples from the Fourier
transform are insignificant - at least in comparison to other systematic
errors and noise. So the worry about truncation is many times unwarranted.
Now one very nice thing about working in G(r) is that it is intuitive and
in real space. The Si-O example was mentioned earlier. Many times one can
get results from simple peak fitting to near neighbor PDF peaks - or study
the width as function of temperatures (as beautifully demonstrated in
Simon Billinge's 1996 PRL). More recently people exploit that one can
restrict the refinement range and look at local geometries (e.g.
Jahn-Teller distortions)  and see where the cross over to the average
behavior is.

So it's cool, it's becoming easier - try your own PDF experiment ;-)
Disclaimer: I am going to hit send without reading what I wrote.

Thomas

-- 
Thomas Proffen
NPDF instrument scientist
Lujan Neutron Scattering Center
Los Alamos National Laboratory
http://www/lansce.lanl.gov/

Reply via email to