[Ifeffit] Consistency criterion

Scott Calvin SCalvin at slc.edu
Thu Sep 28 07:53:00 CDT 2006

Hi Juan,

They're consistent. The higher r-factor on the kwt-1 data might be a 
clue about how reliable part of your data range is...since it's the 
kwt-1 that's higher, it may be that the low end of the k-range is not 
being fit very well (look at the k-space and q-space fits to help 
confirm this). That could be just because of glitchy data, iffy 
background subtraction, or that your model doesn't do as well with 
low-Z scatterers like oxygen as it does with the higher-Z 
contributions (the low-Z scatterers have most of their contributions 
at low k). Aside from looking at the k- and q-space fits, I'd try 
increasing kmin and see if that improves the fit. If it does, you 
might want to do it on the kwt-3 fit too, because that would suggest 
you're not fitting that low-k data very well, and the problem is just 
less emphasized with kwt-3. Of course it would be even better to find 
the source of the problem and address it.

Having said that, I should emphasize that these are non-linear fits. 
Sometimes it's hard to come up with a simple explanation for why one 
fit is closer than another.

Nevertheless, your fits are consistent. They're not equally 
"good"...but that's a different statement. In your situation, I would 
be comfortable publishing the kwt-3 fit and saying that other kwts 
gave consistent results. I would be more comfortable if I could 
figure out what was troubling the kwt-1 fit, though...especially if I 
had a mix of scatterers with substantially different atomic number 
(like transition metals with oxygens).

--Scott Calvin
Sarah Lawrence College

At 08:03 AM 9/28/2006, you wrote:

>Hi all,
>Scott answered my question about consistency but now I have a new doubt.
>I have two fits (one in k1 and other in k3 weights) with fitted
>parameters that have
>uncertainty ranges overlaped, then... in principle, they are consitent,
>But the fit in k1 weight have a R-factor too high (0.06) and therefore
>is not a good fit. So I do not know if I can say that they are
>consistent even so.

More information about the Ifeffit mailing list