[Ifeffit] Fit XANES spectra using Athena

Scott Calvin SCalvin at slc.edu
Sun Oct 26 12:39:47 CDT 2008


I haven't been following this discussion closely, but actually it is  
possible to use PCA to help with a single spectrum of a single sample.  
The "trick" is to use linear combinations of the spectra of the  
standards to create spectra which look like mixtures with various  
amounts of different phases in them. Try to include all the phases  
that you think might be in the sample, except one. Then run the sample  
and the mixtures of standards together through PCA. Finally try a  
target transform to the "missing" phase. If it's in the sample, it  
should work pretty well. As a control, you have to also try PCA with  
the mixtures of standards but with the sample left out, and try the  
same target transform--it should work poorly.

I've used similar tricks before to confirm the presence of certain  
substances in a sample...it can be a very useful technique when you  
can't pin down one of the substances in the sample, but suspect what  
some of the other contributors are.

--Scott Calvin
Sarah Lawrence College

On Oct 26, 2008, at 1:16 PM, Frenkel, Anatoly wrote:

> PCA works only if there are series of spectra with change in the  
> makeup of S species, not for a single spectrum.
>
> Anatoly
>
>
> ----- Original Message -----
> From: ifeffit-bounces at millenia.cars.aps.anl.gov <ifeffit-bounces at millenia.cars.aps.anl.gov 
> >
> To: ifeffit at millenia.cars.aps.anl.gov <ifeffit at millenia.cars.aps.anl.gov 
> >
> Sent: Sun Oct 26 13:05:32 2008
> Subject: Re: [Ifeffit] Fit XANES spectra using Athena
>
> Jenny,
>
> I'm relatively new to the community XAFS, and have not done any  
> sulfur work thus far... but have you considered doing a principal  
> component analysis on the XANES of your samples?
>
> This method should help you identify how many phases are present  
> (assuming that all of your samples aren't exactly identical; however  
> the maximum number of independent phases it will return is the  
> number of samples you tested), and with your standards you might be  
> able to eliminate some of the phases (from the standards) as well as  
> identify some of the phases that are in your sample.
>
> (There is software that will do this in the IFFEFIT package under  
> Sixpack --> 'pc analysis')
>
> Andrew Campos




More information about the Ifeffit mailing list