linear combination fitting
Dear Bruce, I am using Dathena to do some linear combination fitting. In your program, you give three methods for fitting, nomal, derivate and chi(k). I got three different results using these methods. My question is which one is better to obtain more reasonable result or what the purpose of each method is. Regards, Shaofeng
Dear Shaofeng,
On Wed, Aug 7, 2013 at 6:36 AM, Shaofeng Wang
Dear Bruce,
I am using Dathena to do some linear combination fitting. In your program, you give three methods for fitting, nomal, derivate and chi(k). I got three different results using these methods. My question is which one is better to obtain more reasonable result or what the purpose of each method is.
Yes, it is. The other ones are definitely worse. Now, slightly more seriously (yes, that was a joke), how different are the results? how different are fit ranges for these three fits? How good are the data and standards used over these ranges? As you might expect, a linear combination for chi(k) is generally intended for EXAFS, and so emphasizes distances of the nearest neighbors while "norm" and "derivative" is generally for the XANES portion of the spectra, and so emphasizes chemical state. Cheers, --Matt PS: Have you read the postings on this mailing list in the past couple months and weeks? Next time, please try to not exasperate Bruce (and the rest of us), and ask a question that can possibly be answered. Concrete examples and questions are much, much better than abstract questions. Please avoid asking "why is the blue curve always lower than the red curve"?
On 08/07/2013 07:55 AM, Matt Newville wrote:
How good are the data and standards used over these ranges? As you might expect, a linear combination for chi(k) is generally intended for EXAFS, and so emphasizes distances of the nearest neighbors while "norm" and "derivative" is generally for the XANES portion of the spectra, and so emphasizes chemical state.
I want to expand on this point. Have you ever measured data on a metallic nanoparticle? I'll use gold as my example. If you compare good XAS data on gold nanoparticles to data on a gold foil, they are very similar, but not identical. The XANES for the nanoparticles is quite similar -- possibly indistinguishable -- from the XANES for the foil. The EXAFS, however, is quite different due in part to the effect of undercoordination of the scattering shells and in part to the additional structural disorder introduced by the large surface to volume ratio. Now imagine that you are trying to do linear combination fitting on a system containing gold. If you are doing LCF on the XANES, it probably doesn't matter whether you use data on a foil or on nanoparticles as a standard. If you are doing LCF on the EXAFS, it most certainly matters. My point is that, like Matt, I think you asked the wrong question. None of the ways of doing LCF are "better" in some general sense. You need to think hard about the problem you are trying to solve and do the thing that is the most appropriate way to find your answer. LCF using chi(k) is sometimes a useful tool. LCF using norm or deriv is sometimes a useful tool. As for the difference between norm(E) and deriv(E) -- well, I expect that they would give the same answers, within numerical precision and measurement uncertainty, for the same data. Why does Athena allow either one? Well, it's not up to me to choose. It's up to you! HTH, B -- Bruce Ravel ------------------------------------ bravel@bnl.gov National Institute of Standards and Technology Synchrotron Science Group at NSLS --- Beamlines U7A, X24A, X23A2 Building 535A Upton NY, 11973 Homepage: http://xafs.org/BruceRavel Software: https://github.com/bruceravel
participants (3)
-
Bruce Ravel -
Matt Newville -
Shaofeng Wang