number of independent points
Hello, I am wondering what values Artemis/feffit use for kmax and kmin when calculating the number of independent points (Nidp) for a fit done in R-space, where according the feffit document, Nidp = { [2*(kmax - kmin)*(Rmax-Rmin) / pi] + 2}, and kmax and kmin define the fitting range. I have tried using the Fourier transform range for kmax and kmin, but this does not give agreement with the value of Nidp reported by Artermis. Thank you, George
Hello, Since I haven't received a response to my previous question, I just wanted to make sure I asked it clearly. To elaborate, I am attaching a log from Athena. The log file lists: "Independent points = 27.057617188" If I attempt to reproduce this result using the equation in the feffit document, I find Nidp = { [2*(kmax - kmin)*(Rmax-Rmin) / pi] + 2} = { [2*(14 - 0.5)*(4.2 - 1.0) / pi] + 2} = 29.50197417, where I have used the k-range and R-range in the log file as kmax, kmin, Rmax, and Rmin. Can someone explain to me why may calculation does not reproduce the number given for "independent points" in the log fie? Thank you, George On Fri, Oct 7, 2011 at 1:04 PM, George Sterbinsky < GeorgeSterbinsky@u.northwestern.edu> wrote:
Hello,
I am wondering what values Artemis/feffit use for kmax and kmin when calculating the number of independent points (Nidp) for a fit done in R-space, where according the feffit document, Nidp = { [2*(kmax - kmin)*(Rmax-Rmin) / pi] + 2}, and kmax and kmin define the fitting range. I have tried using the Fourier transform range for kmax and kmin, but this does not give agreement with the value of Nidp reported by Artermis.
Thank you, George
Sorry, the log file is from Artemis, not Athena... On Mon, Oct 10, 2011 at 10:35 AM, George Sterbinsky < GeorgeSterbinsky@u.northwestern.edu> wrote:
Hello,
Since I haven't received a response to my previous question, I just wanted to make sure I asked it clearly. To elaborate, I am attaching a log from Athena. The log file lists:
"Independent points = 27.057617188"
If I attempt to reproduce this result using the equation in the feffit document, I find
Nidp = { [2*(kmax - kmin)*(Rmax-Rmin) / pi] + 2} = { [2*(14 - 0.5)*(4.2 - 1.0) / pi] + 2} = 29.50197417,
where I have used the k-range and R-range in the log file as kmax, kmin, Rmax, and Rmin.
Can someone explain to me why may calculation does not reproduce the number given for "independent points" in the log fie?
Thank you, George
On Fri, Oct 7, 2011 at 1:04 PM, George Sterbinsky < GeorgeSterbinsky@u.northwestern.edu> wrote:
Hello,
I am wondering what values Artemis/feffit use for kmax and kmin when calculating the number of independent points (Nidp) for a fit done in R-space, where according the feffit document, Nidp = { [2*(kmax - kmin)*(Rmax-Rmin) / pi] + 2}, and kmax and kmin define the fitting range. I have tried using the Fourier transform range for kmax and kmin, but this does not give agreement with the value of Nidp reported by Artermis.
Thank you, George
Try it again with dk and dR extending the range.
Perhaps there is a more subtle analysis of the data range, but I doubt
it, since Matt is on record as suggesting that people not place too much
emphasis on the exact value of N-idp, if I recall correctly. You can
scale your chi-square, reduced chi-square, and errors using whatever
value of N-idp you prefer.
Jeremy
Chemical Sciences and Engineering Division
Argonne National Laboratory
Argonne, IL 60439
Ph: 630.252.9398
Fx: 630.252.9917
Email: kropf@anl.gov
From: ifeffit-bounces@millenia.cars.aps.anl.gov
[mailto:ifeffit-bounces@millenia.cars.aps.anl.gov] On Behalf Of George
Sterbinsky
Sent: Monday, October 10, 2011 9:39 AM
To: XAFS Analysis using Ifeffit
Subject: Re: [Ifeffit] number of independent points
Sorry, the log file is from Artemis, not Athena...
On Mon, Oct 10, 2011 at 10:35 AM, George Sterbinsky
On Monday, October 10, 2011 10:48:25 am Kropf, Arthur Jeremy wrote:
Perhaps there is a more subtle analysis of the data range, but I doubt it, since Matt is on record as suggesting that people not place too much emphasis on the exact value of N-idp, if I recall correctly. You can scale your chi-square, reduced chi-square, and errors using whatever value of N-idp you prefer.
The papers by Rossner and Krappe (for instance, http://dx.doi.org/10.1103/PhysRevB.66.184303) use a Bayesian approach to evaluate the actual information content in an EXAFS signal. Unsurprisingly, the Bayesian analysis finds something we should already know -- the EXAFS signal is not an ideally packed sum of sine waves. Consequently, the actual information content is rather less than the Nyquist criterion. That is the context of the suggestion not to place to much emphasis on the Nyquist number. (Since Jeremy invoked Matt, I should say that I suspect that Matt would more or less agree with what I am saying here.) It gives us a number to work with in evaluating the statistics in Ifeffit. The interpretation of Ifeffit's statistical analysis (which is a simple Gaussian analysis for a problem that is ill-posed in the Guassian sense) has been a frequent topic of conversation on this mailinglist over the years. It is also a topic in this talk that I sometimes give: https://github.com/bruceravel/XAS- Education/downloads/atea.pdf All that said, the question of the formula used and the value reported in Artemis is a valid one. I don't recall off the top of my head, but I will look at the codes. 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/ifeffit/Demeter
Hi George. It should be 2*(Delta k)*(Delta R), without the +2, but as a floating point number. That is, it is deliberately conservative. I believe the subtle difference you see (that is why the value is 27.06 instead of 27.5) must be due to rounding of the R values on to a discrete grid. -Matt On Oct 10, 2011 9:35 AM, "George Sterbinsky" < GeorgeSterbinsky@u.northwestern.edu> wrote:
Hello,
Since I haven't received a response to my previous question, I just wanted to make sure I asked it clearly. To elaborate, I am attaching a log from Athena. The log file lists:
"Independent points = 27.057617188"
If I attempt to reproduce this result using the equation in the feffit document, I find
Nidp = { [2*(kmax - kmin)*(Rmax-Rmin) / pi] + 2} = { [2*(14 - 0.5)*(4.2 - 1.0) / pi] + 2} = 29.50197417,
where I have used the k-range and R-range in the log file as kmax, kmin, Rmax, and Rmin.
Can someone explain to me why may calculation does not reproduce the number given for "independent points" in the log fie?
Thank you, George
On Fri, Oct 7, 2011 at 1:04 PM, George Sterbinsky < GeorgeSterbinsky@u.northwestern.edu> wrote:
Hello,
I am wondering what values Artemis/feffit use for kmax and kmin when calculating the number of independent points (Nidp) for a fit done in R-space, where according the feffit document, Nidp = { [2*(kmax - kmin)*(Rmax-Rmin) / pi] + 2}, and kmax and kmin define the fitting range. I have tried using the Fourier transform range for kmax and kmin, but this does not give agreement with the value of Nidp reported by Artermis.
Thank you, George
_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit
On Monday, October 10, 2011 10:35:12 am George Sterbinsky wrote:
Since I haven't received a response to my previous question, I just wanted to make sure I asked it clearly.
I suspect the lack of response had more to do with it being a non-trivial question asked on a Friday afternoon :)
To elaborate, I am attaching a log from Athena. The log file lists:
"Independent points = 27.057617188"
If I attempt to reproduce this result using the equation in the feffit document, I find
Nidp = { [2*(kmax - kmin)*(Rmax-Rmin) / pi] + 2} = { [2*(14 - 0.5)*(4.2 - 1.0) / pi] + 2} = 29.50197417,
where I have used the k-range and R-range in the log file as kmax, kmin, Rmax, and Rmin.
Can someone explain to me why may calculation does not reproduce the number given for "independent points" in the log fie?
George, A quick examination of the ifeffit source code shows that there are two issues: 1. Regardless of what it may say in the document, Ifeffit uses "+0" rather than "+2". See line 345 of http://cars9.uchicago.edu/svn/ifeffit/trunk/src/lib/iff_feffit.f A bit of history: Ed Stern wrote this paper back in 1993: http://dx.doi.org/10.1103/PhysRevB.48.9825 Over the years, many have expressed scepticism of his conclusion. The Bayesian work I mentioned earlier suggests quibbling over the value of N in the "+N" term misses the bigger issue. Ifeffit's use of "+0" provides a conservative yet easily stated estimate of Nidp. 2. The difference to the left of the decimal point has to do with the fact that Artemis allows you to pick any value for kmin, kmax, rmin, and rmax. When Ifeffit evaluates Nidp (and anything else) is actually takes the nearest grid points to those four values. The grid in k-space is 0.05 inv. Angstroms. This sets the grid in R space to be about 0.03 Angstroms, but not exactly. So the values of Rmax and Rmin used by Ifeffit to compute Nidp (and other things) are not quite what you specified. At some point, George, you may want to bite the bullet and look at source code. Details are not unknowable, they just might not be written down anywhere beside the source code. Here is the top of Ifeffit: http://cars9.uchicago.edu/svn/ifeffit/trunk/ 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/ifeffit/Demeter
Bruce, Matt, and Jeremy,
Thank you for explaining this. Your responses are much appreciated.
George
On Mon, Oct 10, 2011 at 11:35 AM, Bruce Ravel
On Monday, October 10, 2011 10:35:12 am George Sterbinsky wrote:
Since I haven't received a response to my previous question, I just wanted to make sure I asked it clearly.
I suspect the lack of response had more to do with it being a non-trivial question asked on a Friday afternoon :)
To elaborate, I am attaching a log from Athena. The log file lists:
"Independent points = 27.057617188"
If I attempt to reproduce this result using the equation in the feffit document, I find
Nidp = { [2*(kmax - kmin)*(Rmax-Rmin) / pi] + 2} = { [2*(14 - 0.5)*(4.2 - 1.0) / pi] + 2} = 29.50197417,
where I have used the k-range and R-range in the log file as kmax, kmin, Rmax, and Rmin.
Can someone explain to me why may calculation does not reproduce the number given for "independent points" in the log fie?
George,
A quick examination of the ifeffit source code shows that there are two issues:
1. Regardless of what it may say in the document, Ifeffit uses "+0" rather than "+2". See line 345 of http://cars9.uchicago.edu/svn/ifeffit/trunk/src/lib/iff_feffit.f
A bit of history: Ed Stern wrote this paper back in 1993: http://dx.doi.org/10.1103/PhysRevB.48.9825
Over the years, many have expressed scepticism of his conclusion. The Bayesian work I mentioned earlier suggests quibbling over the value of N in the "+N" term misses the bigger issue. Ifeffit's use of "+0" provides a conservative yet easily stated estimate of Nidp.
2. The difference to the left of the decimal point has to do with the fact that Artemis allows you to pick any value for kmin, kmax, rmin, and rmax. When Ifeffit evaluates Nidp (and anything else) is actually takes the nearest grid points to those four values. The grid in k-space is 0.05 inv. Angstroms. This sets the grid in R space to be about 0.03 Angstroms, but not exactly. So the values of Rmax and Rmin used by Ifeffit to compute Nidp (and other things) are not quite what you specified.
At some point, George, you may want to bite the bullet and look at source code. Details are not unknowable, they just might not be written down anywhere beside the source code.
Here is the top of Ifeffit: http://cars9.uchicago.edu/svn/ifeffit/trunk/
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/ifeffit/Demeter _______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit
participants (4)
-
Bruce Ravel -
George Sterbinsky -
Kropf, Arthur Jeremy -
Matt Newville