Re: GSAS - R(free) sampling factor and seed

[Brian H. Toby]

The idea for R(free) comes from protein crystallography and is a good 
one for folks who have more data than they need. This is commonly true 
of macromolecular work, but this is rarely true in powder work. The gist 
of R(free) is this: rather than look at the agreement between 
reflections *that you have used to fit your model* and the values your 
model predicts, select a random subset of reflections that you *do not 
include* in your fit and see how well they match the model. This is a 
far more rigorous test of a model and thus provides a more unbiased 
agreement factor.

The problem for powder diffraction is that one must exclude fairly wide 
sections of the pattern to drop a reflection from the fit and one needs 
to drop a significant number of reflections over a wide range in Q so 
that R(free) will mean anything. This means that to compute R(free), one 
will end up not using a significant fraction of the data set. Perhaps 
this is not always true, but it seems that in almost every structure 
that I work on, I end up wanting even more data than I have, so that I 
try more sophisticated models -- I'd be very hard pressed to give up 
some data and give up on fitting parameters that I know matter.

I don't know how exactly Bob has implemented R(free) in GSAS, but I 
suspect it is documented in the GSAS manual. I am not sure you need to 
bother looking; I also suspect that R(free) will not be of great value 
unless you have a huge unit cell and very high resolution data or have 
collected data over a large Q range with a very small wavelength.

Brian
[EMAIL PROTECTED] wrote:
Dear Rietvelders,
I am using the updated version of GSAS and will like to know bit more about the
following feature(s)?
In the histogram editing menu, What is the physical meanings of the R-free
sampling factor and seed - (option S) ? How these affect the refinement
result?
Hope that this will not too naive to ask these question in this community..
many thanks,
stephen chui