[sdpd] Meaningless Chi2

[Armel Le Bail <alb...@cristal.org>]

Not convinced ?

Rexp = { (No - Np) / Sum [wi*yi(obs)**2] }**-2

No = number of profile points,
Np = number of refined parameters,

and wi for conventional X-rays and neutron data is
usually assigned :
                               wi = 1/yi(obs)
Then Rexp is really simply :
Rexp = { (No - Np) / Sum [yi(obs)] }**-2
directly dependent of the step by step sum of the pattern intensities.

Since Chi2 depends on the Rwp/Rexp ratio, then

Count more and Rexp decreases, then Chi2 increases.
Count less and Rexp increases, then Chi2 decreases,
because Rwp will not change a lot, certainly not as much
as Rexp,
Count for the right Sum [yi(obs)], or rescale appropriately, and Chi2 =1
exactly as you wish.

Believe me, only trust Rp for qualifying a good whole profile
fit. Preferably obtain Rp < 10%. Because it is too easy to
obtain Chi2 = 1, even with Rp and Rwp values > to 15 or 20%
or even as high as you want.

I hate seeing papers not giving Rp nor even Rwp and Rexp.
Some published papers give only Chi2 close to one (and a
horrible fit drawing which let you suppose that Rp was too
high to be disclosed ;-). This has to be avoided.

Now, if you define wi differently...

Best,

Armel


------------------------ Yahoo! Groups Sponsor ---------------------~-->
FREE COLLEGE MONEY
CLICK HERE to search
600,000 scholarships!
http://us.click.yahoo.com/47cccB/4m7CAA/ySSFAA/UIYolB/TM
---------------------------------------------------------------------~->

 

Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/