[Ifeffit] Guess Parameters Getting Stuck

[Matt Newville]

Hi Aleks,

On Thu, May 25, 2017 at 2:46 PM, M Solo <sarsol89 at gmail.com> wrote:

> Hello.
> I'm currently in process of fitting some exafs data, and have encountered
> this issue several times.
> At some point during fit one or several "guess" parameters get stuck on
> the initial guess value, even if that value is changed to something that
> has no resemblance to the data being fitted. Further changing the initial
> guess value has no effect, and the fit merely takes in the new value as if
> its been "set". The number of values affected range from 1 to all of the
> variables in the GDS.
>


Unfortunately, that's not enough information for us to help you diagnose
the trouble you're having.  As you've probably heard us say many time to
other requests on this mailing list, you'll have to give more details,
preferably a small runnable example that demonstrates the problem.  Athena
and Artemis project files are a good way to do this.

The way fitting generally works, if a variable is not seen to have an
effect on the resulting fit, it's value is not changed.   If, for example,
you guess sigma2 for an EXAFS path to be 100, it will almost certainly
cause the contribution of that path to chi(k) be 0.  Furthermore, a small
change in that sigma2 will not change the fit at all -- the variable will
have been observed to not be important and will not be updated.

Of course, we don't whether such a problem might be happening for you.

--Matt Newville
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://millenia.cars.aps.anl.gov/pipermail/ifeffit/attachments/20170525/c9fb56fc/attachment.html>