Advice on refining GSAS profile coefficients

Date: Tue, 3 Nov 1998 17:24:51 -0500 (EST)
To: RIETVELD_L Distribution List 
From: Srebri Petrov 
Subject: Re: GSAS profile coefficients
Reply-To: RIETVELD_L Distribution List 

At 08:26 PM 11/2/98 EST, you wrote:
>Hi,
>My questions pertain to refining the profile coefficients.  What order do you
>refine the profile coefficients?  Do you refine one, two or more at a time?
>If you refine the profile coefficients in stages do you add to the list or
>toggle the other one off?
>Thanks
>MIZ
>
 Dear MIZ,

The question sounds borring, but is of great importans for everyone, using
GSAS (and probably the for the users of other Rietveld codes). In GSAS we
have 19 profile parameters and it could happen (very often) that you can
screw up the fit that you have been reached just by inprompt final attempt
to improve it, refining one or few of these parameters. Unfortunately, there
is no general rule for the right strategy one should follow. As a first
step, try to understand the meaning of all these parameters, or at least the
first 13 coefficients. Literature: GSAS, Manual, Thompson, Cox & Hastings
and Finger, Cox & Jephcoat, references are provided in GSAS. At the
begining, try to use fixed values for the first 3 of them, related to U, V
and W - use the default values in inst_xry.prm file. Try to refine the
experimental profiles of your pattern, swithcing "on" one by one GP (#4), LX
(#5), and LZ (#6). Open one of them for refining, while all others are
"off". Then, close the refined parameter and open the next one, etc. You
should repeat this few times if the results are not good. Every time, after
runing Genles, you should run Powpref and if the value of the coefficient
has become unreal large, you will get the message. Change it back to 0.0 and
start again. If you are lucky to reach good fit, go ahead, with S/L and H/L
(# 7 and 8) both together). Be sure from the very begining, that your unit
cell parameters are refined and the marks matched well to the peaks
positions - this is very important to be able to control only the profile
fitting. Next two, "trns (# 9) and shft (# 10) depends on chemistry and
geometrical parameters of your Goniometer, so refine them one by one. The
last 3 parameters you can refine again one by one. At the end, you can go
back and refine GU, GV and GW - one by one. The described way is a strategy,
based on the concept not to touch your atomic parameters (x, y, z, ISO and
OCCUP) until you reach the acceptable profile fit. Some people are refinig
all refinable parameters, including the structural ones) at the same time,
but I am not follower of such strategy. 
Another, and probably the best strategy, is to run a standard sample (Si,
LaB6), and to obtain the profile coefficients for your instrument. The yield
values for all 13 parameters you can write in your inst_xry.prm file and use
it for each sample you will refine in the future.
Another, time consuming, but quite useful exersice, is to cut one - two or
three peaks only from a diffraction pattern that you have. It is good if you
choose some partially resolved ka1,ka2 reflections. Try to find out how
these coefficients work on such simple graphics, on a very small numbers of
points. This allows you quickly to "see" what each one of them is doing in
the fitting procedure.
I am not sure if these notes can help you, but anyway - good luck, and have
a fun!

Sr. Petrov

/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\

Dr. Sr. Petrov/PXRD Analyses,
Dept. of Chemistry, U of T,
80 St.George Str., R.417
Toronto ON M5S 3H6, Canada
Tel/Fax: (416)-978-1389.
e-mail: spetrov@alchemy.chem.utoronto.ca