[Ifeffit] How to keep enot, delr, and ss within acceptable limits

[Bruce Ravel]

On Friday 09 January 2009 12:34:01 pm Hashem Stietiya wrote:
> I am currently using ARTEMIS to fit Zn-O, Zn-Zn paths to first and second
> shells, respectively. Many times I would get unacceptable values for enot,
> ss, and delr. My question is: is there a way to tell ARTEMIS that I want
> these parameters to be within acceptable limits? For example, you cannot
> have negative ss, so how can I avoid any fit result that produces negative
> values for ss? Or high extremely high values for enot and delr?

Hashem,

There are two possible answers to your question -- the one you are probably 
looking for and the one that involves me getting up on my soapbox and 
lecturing.  So.... in that order....

You are probably looking to use restraints.  If you search the mailing list 
archives, you'll find some discussion of that topic.  You might also find my 
explanation in this presentation useful: 
http://xafs.org/Workshops/APS2008?action=AttachFile&do=get&target=Ravel_Advanced.pdf

In short, a restraint is a term that you add to the evaluation of the fitting 
metric which imposes a penalty on the fit when a parameter wanders away from 
a value that you have prior knowledge about.  The next to last slide in that 
talk shows a screen shot that should make it clear how they get implemented 
in Artemis.

I would caution you, though, against disregarding a fit that results in 
unphysical parameter values.  That a parameter takes on an unphysical value 
is usually telling you something about your fitting model.  If your delta R 
parameter is really large, perhaps your starting structure is not a very good 
representation of your real sample.  If your sigma^2 parameter is negative, 
perhaps you have too much structural disorder in your model and the fit is 
compensating by inflating those paths with a negative sigma^2.

Ifeffit and Artemis do not report things like negative sigma^2 parameters just 
to mess with your head.  (Although that *would* be fun...!)  The parameters 
reported are the parameters required to minimize the fit.  Ifeffit doesn't 
know anything about science -- it just does what it can to make the fitting  
metric as small as possible.  It's kinda stupid that way and Artemis isn't 
any smarter.  Fortunately, _you_ are smarter.  When a physically unreasonable 
parameter is reported, you can examine the assumptions made in your fitting 
model.  If parameters are coming back in an unphysical state, it is almost 
certain that there is something unphysical about your fitting model.

Thus, wonky values for the parameters of a bad fit are full of information 
that you need to interpret and use to refine your fitting model.

Hope that helps,
B


-- 

 Bruce Ravel  ------------------------------------ bravel at bnl.gov

 National Institute of Standards and Technology
 Synchrotron Methods Group at NSLS --- Beamlines U7A, X24A, X23A2
 Building 535A
 Upton NY, 11973

 My homepage:    http://xafs.org/BruceRavel
 EXAFS software: http://cars9.uchicago.edu/~ravel/software/exafs/