As far as I know, GSAS does use a "trial" structure as its starting point for
the GSAS variant of "Le Bail" intensity extraction (ref: Le Bail et al, Mater.
Res. Bull, 1988, 41, 117). I am not certain of this, because I have not looked
at that section of the code. If so, this is not a good practice for ab initio
structure determination as Armel correctly points out, unless you really have a
truly random trial structure (which requires lots of atoms).
One cure for the problem would be a short program that reads the [binary]
reflection file and rewrites it with pseudorandom F(obs) values.
There are some advantages, though for the method used in GSAS -- it can be
helpful for fitting an impurity phase with preferred orientation (as happens
here when we use Al cans or don't mask properly) or when computing Fourier maps
where one has a structure that is partly correct.
A related GSAS tip: when performing a Le Bail extraction and refining
parameters such as peak shapes, etc. Run GENLES a few times with the number of
cycles set to zero to optimize the structure factors before doing any
least-squares cycles. The structure factors are fit every time GENLES is run,
even if no cycles are run. Damping is also often needed, too.
Brian
--***************************************************************************Brian H. Toby Reactor Radiation Division E151/235Brian.Toby@NIST.gov National Institute of Standards & Technologyvoice: 301-975-4297 FAX: 301-921-9847 Gaithersburg, MD 20899***************************************************************************