[Ifeffit] Spine Clamps and Some Fit Results

Matt Newville newville at cars.uchicago.edu
Thu Jan 29 12:56:48 CST 2004


Hi Dan,

On Wed, 28 Jan 2004 dmc at pdx.edu wrote:

> (1) In Athena what exactly do the varying levels of spine clamps mean?  

They mean how strongly to try to make the background mu0(E) match
the value of mu(E) at the endpoints.  Athena uses the qualitative
terms 'strong', 'rigid', and so on because it's difficult to give
physical meaning to the weighting factors used for this portion of
the fit.

> When do I need to use them?  

If the background is clearly diverging from mu(E) at high-k, a clamp
is a useful thing to use.  At low-k it's harder to tell what the
background function is supposed to do, so it's even more of a
judgement call of when to use the low-k clamp.

> I've been ignoring them up to now, but I think that it might help 
> my resulting fit, which brings me to my second part.  

Hopefully, a spline clamp will not greatly alter fitting results.  
The main way the background removal _should_ affect fit results is
how the normalization is done, and what the edge step gets set to.
  
> (2) By looking at my results is there any advice you can give me?  
> Do my reduced chi-squared, R-factor and correlated variables look
> good?  I know that the R-Factor should be below 0.05, but what
> about red chi-squared (depends on the material)?

>From what the limited information you posted, it looks OK.  An
R-factor > 0.05 often indicates a bad fit, but that value is not
case in stone.  The values of chi-square and reduced chi-square are
complicated by depending on the estimated uncertainty in the data
which is an estimate.  Statistics 101 says that reduced chi-square
should be ~1 for a 'good fit', but Statistics 101 assumes that a)
you know the uncertainty in the data, b) you know how many
independent measurements (ie, how much data) you have, and c) the
model has no error in it.  For EXAFS, these can all be questioned.
Experience says that a reduced chi-squares around 10 are common for
excellent fits to data of well-characterized standards, such as a
metal foil or powdered metal oxide.

> Also, I am concerened that my amp (SO^2) should be closer to 1.

Sorry to be so dense, but is S02 set as 'amp' (which is restrained
somehow, and ending up at 0.707), 'ampg', or 'ampt'?  The standard
answer is that S02 between 0.7 and 1.0 is 'normal'.  To say any
more, you'd want to start checking the normalization and energy
resolution.  It might be worth thinking about how S02 is restrained
(if it is), and why the uncertainty in ampg is so small.

Hope that helps!

--Matt






More information about the Ifeffit mailing list