> >Yes, it was done (see J. Appl. Cryst. 26 (1993) 97-103) and the
> >column-length distribution function appears to have an "acceptable"
> >asymmetric shape for tested samples.
>
> You mean that whatever a Voigt function, it corresponds to a
> reasonable size distribution function ? This is hard to believe
> at least when that Voigt function tends towards a pure Gaussian,
No, but the paper discusses exactly that case and cites the physical reason
why it may occur (incorrectly determined background, that is too high, which
often happens because of peak overlapping).
> which is one possibility. Moreover, a Voigt function cannot cover
> all the possibilities for a size distribution. So, what will be the
> error when trying to fit a non-Voigt by a Voigt ?
>
> I am afraid that using analytical shapes will always be contestable.
> They restrain the experimental field to some defined models,
> inevitably. Finally, we can just discuss about which model
> could be the less bad (giving the best fit). The fits in my report
We completely agree on this Armel. As I said in a previous message, for me,
the art is to strike a reasonable compromise between a "perfect" (and still
nonexisting) profile and a reasonably good approximation. I think that your
model belongs to that category. I suggested the addition of a Gaussian to
AnS because it would make it a bit more general at minimal (programming)
cost, but for most cases a single Lorentzian works very well. I guess,
that's why we are doing the round robin, to straighten out these "little"
problems:-)
But now I should go back to analyzing the round-robin results or we'll never
learn the answer:-)
All the best, Davor
