[Ifeffit] Difference File Fits

[Matt Newville]

Hi Norbert, Bruce,

> On Tuesday 18 March 2003 06:56 am, Norbert Weiher wrote:
> 
> > I was just playing a bit around with some data and it sprang to my mind that 
> > sometimes the idea of doing difference file spectra fits would be a nice 
> > feature (like XDAP has it). This would be great to identify single 
> > contributions of certain scatterers that are close to each other or somehow 
> > other produce similar paths.
> > 
> > How about including a feature which lets you extract a "new" chi.dat file from 
> > the original data minus some paths one has selected? This file could then be 
> > further processed.
> 
> Subtracting a sum of paths from the data is pretty easy at the level
> of ifeffit's math expressions.  A feature for subtracting an arbitrary
 
Like Bruce said, subtracting a partial fit (say, of the first
shell) from data, and fitting the residual is possible with
ifeffit. It might even be called easy.  Similarly, working
with the differences between data files is not hard.  I'm not
sure what XDAP does...

Anyway, fitting and then subtracting the contribution from a
set of paths (eg, the first shell) so that you can analyze
further paths is identical to first fitting the set of paths,
then fixing all the parameters to the best fit values to
analyze the higher shells.  Doing the subtraction emphasizes
the rest of the spectra, which can definitely help
visualization of the fit to the higher shells.

But doing the subtraction does come at a price though.  What
you're subtracting is 'only' the best-fit, not necessarily the
truth and definitely not including the uncertainty in the fit.

At the end of the analysis, it's usually best to do a full fit
(that is, with all parameters floating) to understand the
correlations between variables.  Subtracting off a shell makes
that impossible, so whiile analyzing differences has definite
advantages, it should be done with caution.

It is very difficult to subtract uncertainty.

--Matt