[Ifeffit] Fitting Problems via Artemis

Weiher, Norbert norbert.weiher at chem.ethz.ch
Fri Sep 24 16:11:10 CDT 2004


Dear Yu-Chuan,

I quickly comment on the last two points you mentioned in your message.

Concerning the DW factor, you are right that it accounts for vibrational disorder in the sample (in the ideal case). If, however, your material has a high structural disorder by itself, DW factors might become high due to this. The general behaviour is that they should increase with temperature of the measurement. A negative value is physically not possible, and usually means that something is wrong with your fitting path. Possible things to check are
- did you use proper reference paths (i.e. were they correctly computed by feff)
- is it maybe another scatterer which contributes to the EXAFS at this part.

Concerning fit stabilities:
I never encountered the problem you mentioned, i.e. when I use the same starting values for a fit (even for a bad one), the results are the same. A good fit should by definition converge to the same fitting parameters even when starting from slightly different points on the chi2 surface. However, it is possible that you obtain different results from different starting values. Possible things to do if your fit is not stable:
- dont fit all parameters at one run (at least not in the first place)
- start fitting all parameters independently, then grabbing the values of the former fits and slowly approach a stable point in the fit
- if you have a stable result, try freeing all parameters and see if you reach the same point again. This is a quite good indication that you are on a good way.
- if you include another shell, fix all first shell parameters and just fit the second one. Afterwards, vary all parameters again.

Finally, the background refinement is a feature of ARTEMIS to correct the background spline you are using. Unfortunately, I have not much experience with it, so I leave it to the other IFEFFITers in the world to comment on this...

Hope my suggestions help you to have more productive sessions with these great software packages.

Cheers,
Norbert




More information about the Ifeffit mailing list