L. Keller wrote :
>Aren't f an g both analytical functions or am I missing somehing here?
>Fourier (Warren-Averbach) analysis does not assume any profile shape. That is
>why it's often so cumbersome. As soon as one assigns a function to the
>profile, the plot of Fourier coefficients is always smooth and beautiful (but
>questionable, of course). Doesn't simplification already set in at this point
>if one goes the Rietveld route?
The Fourier route is already oversimplification. The Fourier
coefficients are supposed to correspond to size/microstrain
effects in a way which can only be "true" if the sample is homogeneous
and present quite small distortions.
And you miss something because some Rietveld programs
do not use what is usually called "analytical profiles". They may
use "learnt profiles" either in the form of Fourier coefficients
or differently (see for instance the XRS82 and ARIT programs).
About a model for the f part, why unphysical meaning like
negative proportion of columns of n cells should be allowed ?
So that, yes, the model is usually free of the problems found
by a second derivative of experimental Fourier coefficients.
And I think that this is another good idea.
I would be interested in knowing your opinion about the "fundamental
parameters" approach, which describes profiles by analytical
profiles (generally convolutions of squared Lorentzians plus
some artifacts for creating asymmetry), anyway.
Best,
Armel Le Bail
http://sdpd.univ-lemans.fr/course/
