Sorry, Bruce. I have to send it again to the group. -Jenny
--- On Sun, 10/26/08, Jenny Cai
Hello everyone,
Sorry for bothering you again.
I am using least-squares peak fit and linear combination fit to analyze my samples. I have spent tons of time on it, and it really makes me crazy. Why can't I get consistent results from these two methods?
Please see the attached file. Both of these methods work well individually, but linear combination fit always need more peaks than peak fit to get an 'ok' fitting. Should I stick on one method for all my samples, no matter what results the other one gives? It is confusing me so much. Could anyone help me out?
Jenny, I think you are comparing apples and oranges. I certainly think you are looking for some correspondance between peak fitting and linear combination fitting that probably does not exist. When doing linear combination fitting, you are making the implicit assumption that your data can be described in that way. Suppose that you go to the chemical cabinet and grab jars of 5 stable, non-reactive chemicals. Scoop out a spatula-full of each and mix them together very well in ajar. Then scoop out a bit of the mixture, spread it on a piece of tape, and measure some XAS. In that case, we certainly expect LCF to work well. The measured spectrum should be a linear combination of the spectra from each of the 5 original materials in proportion that has something to do with how many absorber atoms were in each scoop. Now suppose that you take a scoop of soil from a swamp. You might expect that your metal atom is distributed among oxide and sulfide species. You measure some xas on the swamp soil and on a library of standards. You may find that you can do LCF using some mixture of oxides and sulfide, or you might not. That may mean that you neglected to consider an important standard, a carbide, for example. *Or* it might mean that the oxides and sulfides in the real sample include some kind of exotic organic species that isn't quite like the standards you have available. In the latter case, your LCF fit results will be approximate in the best scenario. So, you don't say anything about what sulfur species you are using as standards, nor do you say anything about what the sample is. Should you expect a perfect fit from some number of standards? Do you know for a fact that your standard is simply a linear combination of standards. That is, should you expect the LCF to tell you what the components *are* or should you expect the LCF to tell what the sample resembles? As for the peak fitting, what do each of the line shapes mean physically? Chemically? No XAS data is described by a single Gaussian. Neither is any XAS data described by a single arc-tangent. So why should the number of lineshapes required to generate a mathematical function that resembles your data be related to the number of species you require in a linear combination fit? You seem to making yourself crazy based on the assumption that there is some magical relationship between the six lineshapes you used in the peak fitting and the number of standards you use in the linear combination fitting. It seems to me that you would make yourself a lot less crazy if you didn't cling to this unsubstantiated assumption. 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/ _______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Now with a new friend-happy design! Try the new Yahoo! Canada Messenger __________________________________________________________________ Yahoo! Canada Toolbar: Search from anywhere on the web, and bookmark your favourite sites. Download it now at http://ca.toolbar.yahoo.com.
Hi Jenny,
I understand the difference between peak fitting and linear combination fitting. It is like, If I don't know anything about my sample, peak fitting should be used. If I know the components of my sample, I can use linear combination. That is why I use peak fitting first, because I know very little about my sample. However, what I am thinking is that from the results of peak fitting, I know there may be some types of sulphur in my sample, and based on the peak positions, I can chose the standards for linear combination fitting.
I don't think the two methods are normally used together but rather are used to get completely different kinds of information. As Bruce mentioned, there is no relationship between the results of one in relation to the other.
I can't use linear combination fitting in the first place, because I don't exactly know what sulphur forms in my sample, and I may miss an important species if I chose the wrong standards. However, I still want to try linear combination after peak fitting, because I think the linear combination fitting may be more accurate than peak fitting. It uses the whole spectrum of the standard, while the peak fitting neglectes the effect of small features on the fitting. Does it make sense?
I am not sure what you mean by 'more accurate'. I suppose it depends on how you define accuracy. I hope this this doesn't sound banal but the basic principle of both peak fitting and linear combination fitting is the same. You are trying to use some basis set of functions to model a spectrum. A point that I think Bruce was making is that the important part is in how you interpret the functions. Obviously, a collection of mathematical functions has no physical meaning in terms of species. On the hand, if you just randomly combine some spectra from a standard list that give you a nice fit it does not have any more physical meaning that the Gaussian functions.
My samples are fluid coke and their activation products. They have reduced and oxidized sulphur species, more likely in organic forms. I have 26 standards covering the sulphur oxidation states from elemental sulphur to sulphate. Most of them are organic sulphur.
Any more suggestions? Thank you!
Best regards,
Jenny
I hesitant to mention but I suppose you could take a brute-force approach and fit many combinations of your standards. The result though is that you will likely find multiple combinations that all add up to fit your spectrum with equal statistical goodness of fit. You then have to decide which (if any) might be the right one. You need chemical information to either know what is in your sample or at least what is most likely. You could then perhaps narrow down the search space to find something that is consistent with the data. For example if you happen to have a species with a particularly unique signature. However, there are a large number of organic sulphur species so this seems unlikely. Rather than trying to fit specific species you might try to see if you can classify groups of related sulphur compounds. Can you match peaks in your spectra with major features of a class of sulphur species? In other words, using vibrational spectroscopy terms, look for 'functional groups' first. Once you have separated groups of candidates then use chemical or other information to narrow the list. You might not be able to identify exact species but that may not be necessary. I have no idea what you what to know but do you need to know the exact composition? Or only the types of products? Or how reduced/oxidized the sample is? Inorganic/organic sulphur? You may be able to answer these sorts of questions without knowing the exact composition. I hope this helps. Cheers, Adam
Jenny,
Here are some links to the work of A. Vairavamurthy who did extensive
research on sulfur speciation, that may help:
HYPERLINK
"http://www.irnase.csic.es/users/delrio/repository/2002-OLIVELLA-JAAP-63-59.
pdf"http://www.irnase.csic.es/users/delrio/repository/2002-OLIVELLA-JAAP-63-
59.pdf
HYPERLINK
"http://www.irnase.csic.es/users/delrio/repository/2002-OLIVELLA-FUEL.pdf"ht
tp://www.irnase.csic.es/users/delrio/repository/2002-OLIVELLA-FUEL.pdf
HYPERLINK
"https://dspace.ucalgary.ca/bitstream/1880/44981/1/Jalilehvand_2005_J0007.pd
f"https://dspace.ucalgary.ca/bitstream/1880/44981/1/Jalilehvand_2005_J0007.p
df
The good resource for many issues on sulfur speciation are original articles
by Pickering (1998) and Vairavamurthy (1998).
I think, putting it conservatively, you can use peak fitting to find the
mixing fractions of each charge state of S, and you can use LCF to get a
better idea of what functional groups each is in, which is obviously not the
same (several functional groups may have S in the same state).
Anatoly
From: ifeffit-bounces@millenia.cars.aps.anl.gov
[mailto:ifeffit-bounces@millenia.cars.aps.anl.gov] On Behalf Of Jenny Cai
Sent: Sunday, October 26, 2008 2:12 AM
To: XAFS Analysis using Ifeffit
Subject: [Ifeffit] Fw: Re: Fit XANES spectra using Athena
Sorry, Bruce. I have to send it again to the group. -Jenny
--- On Sun, 10/26/08, Jenny Cai
Hello everyone,
Sorry for bothering you again.
I am using least-squares peak fit and linear combination fit to analyze my
samples. I have spent tons of time on it, and it really makes me crazy.
Why
can't I get consistent results from these two methods?
Please see the attached file. Both of these methods work well
individually,
but linear combination fit always need more peaks than peak fit to get an
'ok' fitting. Should I stick on one method for all my samples, no
matter
what results the other one gives? It is confusing me so much. Could anyone
help me out?
Jenny, I think you are comparing apples and oranges. I certainly think you are looking for some correspondance between peak fitting and linear combination fitting that probably does not exist. When doing linear combination fitting, you are making the implicit assumption that your data can be described in that way. Suppose that you go to the chemical cabinet and grab jars of 5 stable, non-reactive chemicals. Scoop out a spatula-full of each and mix them together very well in ajar. Then scoop out a bit of the mixture, spread it on a piece of tape, and measure some XAS. In that case, we certainly expect LCF to work well. The measured spectrum should be a linear combination of the spectra from each of the 5 original materials in proportion that has something to do with how many absorber atoms were in each scoop. Now suppose that you take a scoop of soil from a swamp. You might expect that your metal atom is distributed among oxide and sulfide species. You measure some xas on the swamp soil and on a library of standards. You may find that you can do LCF using some mixture of oxides and sulfide, or you might not. That may mean that you neglected to consider an important standard, a carbide, for example. *Or* it might mean that the oxides and sulfides in the real sample include some kind of exotic organic species that isn't quite like the standards you have available. In the latter case, your LCF fit results will be approximate in the best scenario. So, you don't say anything about what sulfur species you are using as standards, nor do you say anything about what the sample is. Should you expect a perfect fit from some number of standards? Do you know for a fact that your standard is simply a linear combination of standards. That is, should you expect the LCF to tell you what the components *are* or should you expect the LCF to tell what the sample resembles? As for the peak fitting, what do each of the line shapes mean physically? Chemically? No XAS data is described by a single Gaussian. Neither is any XAS data described by a single arc-tangent. So why should the number of lineshapes required to generate a mathematical function that resembles your data be related to the number of species you require in a linear combination fit? You seem to making yourself crazy based on the assumption that there is some magical relationship between the six lineshapes you used in the peak fitting and the number of standards you use in the linear combination fitting. It seems to me that you would make yourself a lot less crazy if you didn't cling to this unsubstantiated assumption. 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/ _______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit _____ Now with a new friend-happy design! Try the new HYPERLINK "http://ca.beta.messenger.yahoo.com/" \nYahoo! Canada Messenger _____ Image removed by sender.HYPERLINK "http://ca.toolbar.yahoo.com/"Yahoo! Canada Toolbar : Search from anywhere on the web and bookmark your favourite sites. Download it now! No virus found in this incoming message. Checked by AVG. Version: 7.5.549 / Virus Database: 270.8.3/1747 - Release Date: 10/26/2008 9:27 AM No virus found in this outgoing message. Checked by AVG. Version: 7.5.549 / Virus Database: 270.8.3/1747 - Release Date: 10/26/2008 9:27 AM
participants (3)
-
Adam Webb -
Anatoly Frenkel -
Jenny Cai