Athena (linear combination fitting)
Dear Dr. Ravel, Thanks so much,for information you gave me. I subscribed already in the mailing list. I have a complex system to fit. I have a complex spectra from a hetrogenous sample and a number of reference samples. I am doing EXAFS LCF. To choose from the combinations after fitting is diffcult because the R factor of the first 5 combinations with diffrent reference combinations is similar. And as I see from my fitting the alignement plays really an important role on the LCF. You know when I made allign marked groups based on the standard, it alligns well the enregy but when I see the k-plot, it shifts. And my fitting space is on chi(k) which is sensitive when I allign. It is complex system. The procedures I am trying are, Normalization-allignement-LCF (this procedure alligns the energy well but the chi shifts and this affects my fitting because I am fitting on chi space. The other procedure i checked, Normalization-calibration-LCF (this alligns the energy good but for some spectra it shifts again). When I allign it manually by excel sheet, it works good. Can I use the following procedure...normalization, calibration-allignemet-LCF? I just found a well alligned spectra in E and K plot from this procedure. I will be happy if you comment on those procedure. Thanks so much for the keen help! With best regards, Belay
On Thursday, May 12, 2011 03:09:12 am Dilnesa, Belay wrote:
Dear Dr. Ravel, Thanks so much,for information you gave me. I subscribed already in the mailing list. I have a complex system to fit. I have a complex spectra from a hetrogenous sample and a number of reference samples. I am doing EXAFS LCF. To choose from the combinations after fitting is diffcult because the R factor of the first 5 combinations with diffrent reference combinations is similar. And as I see from my fitting the alignement plays really an important role on the LCF. You know when I made allign marked groups based on the standard, it alligns well the enregy but when I see the k-plot, it shifts. And my fitting space is on chi(k) which is sensitive when I allign. It is complex system. The procedures I am trying are, Normalization-allignement-LCF (this procedure alligns the energy well but the chi shifts and this affects my fitting because I am fitting on chi space. The other procedure i checked, Normalization-calibration-LCF (this alligns the energy good but for some spectra it ! shifts again). When I allign it manually by excel sheet, it works good. Can I use the following procedure...normalization, calibration-allignemet-LCF? I just found a well alligned spectra in E and K plot from this procedure. I will be happy if you comment on those procedure. Thanks so much for the keen help!
Hi Belay, LCF fitting to chi(k) spectra is sensitive to choice of E0. This can be quite challenging in a system with large E0 shifts between species. That is, if the oxide and the zero-valent metal are both present in the unknown data, then dealing with the low-k portion of the data can be tricky for a number of reasons. You only get to do a single background removal of the unknown data (obviously!) and how to do the background removals for the pure-phase standards in a way that is consistent with the background removal of the unknown is non-obvious. For this reason, it is a good idea to begin the LCF fit at a fairly large k value -- something like 3 or 4 -- to avoid the systematic uncertainties associated with the background removals. That said, I think your procedure is reasonable, although I am bit unclear what you are doing in your "calibration" step. Are you setting all your data and standards to have the same E0? For LCF analysis, that is probably a reasonable way of dealing with the background removals in the presence of valence edge shifts. However, it is hard for me to comment more clearly without seeing your data. BUT! There is a much better way of convincing yourself that your analysis is being done in a defensible manner than asking the opinion of an "expert". Do an experiment! As I recall from your earlier email, you are working on iron compounds. OK. Prepare a series of artificial samples containing the Fe standards that you are using to interpret your real iron data. Mix the standards together to form a set of combinations. Go to the synchrotron and measure spectra on the mixtures and apply your analysis procedure. If you get the right answers, then you might reasonably have confidence in your analysis procedure. If you get the wrong answer, then you can reassonably question either your procedure or the quality of the software that some clown wrote. HTH, B -- Bruce Ravel ------------------------------------ bravel@bnl.gov National Institute of Standards and Technology Synchrotron Methods Group at NSLS --- Beamlines U7A, X24A, X23A2 Building 535A Upton NY, 11973 My homepage: http://xafs.org/BruceRavel EXAFS software: http://cars9.uchicago.edu/~ravel/software/exafs/
Just to re-iterate and put more bluntly some of what Bruce has said: Doing Linear combination fits of EXAFS chi(k) *requires* that the 1. mu(E)->chi(k) transformation uses the same absolute value for E0 for all spectra. 2. the mu(E) data are on the same energy scale to begin with (or point 1 is impossible). Shifting the energy scales by "align"ing with Athena may be the right thing to do, or it might completely destroy your data, depending on how the data were collected. As linear combination fits (either for XANES normalized mu(E) or EXAFS chi(k)) always give an answer, they are further complicated by the choice of good "standards", and without these can give completely wrong results. Linear combinations fits are more challenging for EXAFS than for XANES, as EXAFS are complicated by disorder terms that do not strongly effect XANES, and this disorder is a "component" of the data. Unless you understand the set of possible atomic environments very well and are only interested in the fractions of known phases, I'd recommend against using simple linear combinations of EXAFS for anything except exploratory analysis. That's all to say that you have to be really careful with these, and you convince yourself that you can extract reasonable results from spectra with known combinations before applying this to unknown data. --Matt
On Thursday, May 12, 2011 10:11:20 am Matt Newville wrote:
Linear combinations fits are more challenging for EXAFS than for XANES, as EXAFS are complicated by disorder terms that do not strongly effect XANES, and this disorder is a "component" of the data. Unless you understand the set of possible atomic environments very well and are only interested in the fractions of known phases, I'd recommend against using simple linear combinations of EXAFS for anything except exploratory analysis.
I certanly didn't make this point strongly enough. I too am wary of chi(k) LCF fits in most situations. A common example of the disorder problem is a system that forms metal nanoparticles. Using bulk metal as a standard in the XANES LCF is much less wrong than using bulk metal as a standard in chi(k) LCF. B -- Bruce Ravel ------------------------------------ bravel@bnl.gov National Institute of Standards and Technology Synchrotron Methods Group at NSLS --- Beamlines U7A, X24A, X23A2 Building 535A Upton NY, 11973 My homepage: http://xafs.org/BruceRavel EXAFS software: http://cars9.uchicago.edu/~ravel/software/exafs/
participants (3)
-
Bruce Ravel
-
Dilnesa, Belay
-
Matt Newville