Re: [Ifeffit] ifeffit R-factor
Hi Sandra,
On Sat, Apr 6, 2013 at 12:37 PM, Sandra Luber
Dear Matt Newville,
I do some fitting with ifeffit using EXAFS data generated by feff. I wonder how the R-factor is calculated. Unfortunately, I have not found any definition yet. Would it be possible that you write me how it is obtained? This would be great.
Thanks a lot.
Best regards Sandra Luber
The basic definition is R = Sum( |data - fit|^2 ) / Sum( |data|^2 ) The old Feffit document probably has the clearest definition, in its Chapter 5 (pages 16-20 of http://cars.uchicago.edu/~newville/feffit/feffit.pdf ). The definition and briefer explanations are also given in several of the tutorials on http://xafs.org/Tutorials. Thanks for reminding me to put this and related information into the Larch documentation! --Matt
Dears, there are already few posts about the doping, simulation of the doping and showing how to prepare the feff.inp file for doped material, but I would like to ask about the opposite situation for low doped materials - how to calculate the doping level from feff.inp file (from the list of ions). If there is already such problem discussed, then please excuse me for the discrepancy and send me the link (I couldn't find it). Just few examples: I construct the feff input file of the radius of 8A (FMS 8.0) and with the SCF of 7. I simulate the spectra with Jfeff9. Let it be Al doped ZnO - simulation on Al:K edge. 1) I investigate the change of the Al:K edge XANES spectra for the situation where I have Al in the centre and vary the distance to the next Al. At the beginning I have only 1 Al in the centre. The 8A radius consist 177 atoms: 86 Zn, 90 O and of course 1 Al. Can I simple say that my doping level in this situation is 1/177=~0.56 at% doping? But this could be also 1/88 if I will reduce the FMS to 6.2A, and the calculation will give the same result. Is any physical limitation to this calculation of doping level? 2) OK, lets add now another Al in the nearest Zn sphere - the XANES spectra changes, as I expected. The doping level doubled from 1/177 to 2/177=~1.13 at%. However, with the shifting of the second Al far away from the 1st Al, I do not observe the change of the XANES spectra, up to 4th sphere. With the 5th sphere (which is at ~6.5A) I can observe the tendency of the change of XANES spectrum toward this in point 1 (probably due to the limited size of the cluster). This means that with the change of the position of the second Al, I change the free means path from ~3.22A for 1st Zn sphere to ~6.14A for the 4th, this could be recalculated as the different doping level from ~5 at% to ~2 at%. Of course following the calculation from the point no. 1 we have in both cases ~1.1 at%... My impression is that I should think in the way of the free means path, but maybe I'm wrong? - I would like to know your attitude. Thanks and best regards kicaj
On Wednesday, April 17, 2013 09:45:40 AM Dr. Dariusz A. Zając wrote:
1) I investigate the change of the Al:K edge XANES spectra for the situation where I have Al in the centre and vary the distance to the next Al. At the beginning I have only 1 Al in the centre. The 8A radius consist 177 atoms: 86 Zn, 90 O and of course 1 Al. Can I simple say that my doping level in this situation is 1/177=~0.56 at% doping? But this could be also 1/88 if I will reduce the FMS to 6.2A, and the calculation will give the same result. Is any physical limitation to this calculation of doping level?
I think there is a small problem is what you wrote here. The cluster you describe is consistent with 0.56 at% doping but also with anything smaller. A more pedantically correct way of describing the cluster you contrusted is that it is consistent with any physical situation in which the Al dopants are, on average, separated by 8 A or more. That necessarily precludes any situation in which there is Al clustering of any sort. If you want to consider the effect of clustering on the XANES calculation, you need first to consider the convergence of the calculation of the host (albeit with the Al absorber) in cluster size. That's not quite the same thing as the mean free path, although there must be some relationship between the two. First you have to figure out convergence as a factor of cluster size. That is, make a calculation with FMS=4.0. Then make one with FMS=5.0. Then FMS=6.0. Keep going until the calculation stops changing. That gives you a sense of what the sensitivity of the calculation could possibly be to unclustered dopants. That is, if the convergence radius is 7.0 A, then the calculation is unlikely to be sensitive to the presence of an Al atom at a diatance of 8.0 A. With low doping levels that, on the average, result in more distant Al atoms, I would expect the computed spectra to be identical.
2) OK, lets add now another Al in the nearest Zn sphere - the XANES spectra changes, as I expected. The doping level doubled from 1/177 to 2/177=~1.13 at%.
That's not what that means. What you have done is to introduce clustering. You are now saying that Al atoms are paired, not that they exist in higher concentration. They may also exist in higher concentrarion. But, strictly speaking, that's not what the addition of an Al in that place means.
However, with the shifting of the second Al far away from the 1st Al, I do not observe the change of the XANES spectra, up to 4th sphere. With the 5th sphere (which is at ~6.5A)
OK. This is the follow-up to the point I made above. The convergence test on the host material sets an upper bound on how far away the second Al atom can be and still be visible in the calculation. You are now probing that with an actual placement of the Al atom.
I can observe the tendency of the change of XANES spectrum toward this in point 1 (probably due to the limited size of the cluster). This means that with the change of the position of the second Al, I change the free means path from ~3.22A for 1st Zn sphere to ~6.14A for the 4th, this could be recalculated as the different doping level from ~5 at% to ~2 at%. Of course following the calculation from the point no. 1 we have in both cases ~1.1 at%...
My impression is that I should think in the way of the free means path, but maybe I'm wrong? - I would like to know your attitude.
MFP is relevant, but it is just a way of putting a number to a more complicated effect. The bottom line is that you want to understand spectral trends as changes are made to the configuration used in a Feff claculation. To do that, you have to make a lot of Feff calculations. 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 Homepage: http://xafs.org/BruceRavel Software: https://github.com/bruceravel
I think there is a small problem is what you wrote here. The cluster you describe is consistent with 0.56 at% doping but also with anything smaller. A more pedantically correct way of describing the cluster you contrusted is that it is consistent with any physical situation in which the Al dopants are, on average, separated by 8 A or more. That necessarily precludes any situation in which there is Al clustering of any sort. yes, actually there is no difference in result between 6A and 8A (and of course large too) clusters. Each enlargement of the cluster size causes
Dear Bruce, Dear Matt, thank you for answers. Short comments to yours: the decrease of the doping level. Actually this was the idea of simulation - observation if an effect of clustering can be observed. Maybe simply we are on the lowest doping level possible to examine, including also the possible detection limits? I can also imagine that below some lowest doping level should not be observed any change in spectra...
If you want to consider the effect of clustering on the XANES calculation, you need first to consider the convergence of the calculation of the host (albeit with the Al absorber) in cluster size. That's not quite the same thing as the mean free path, although there must be some relationship between the two. Of course it has been done at the beginning. The cluster size of 6A for ZnO reached already convergence. The trick is that I see difference in spectra between this with only 1 ion, as an absorber and this with 1 additional, but no difference in distance between Al... I expected that with increasing of the distance between Al I should observe the change of the spectrum toward this with only 1 Al in the centre. That's not what that means. What you have done is to introduce clustering. You are now saying that Al atoms are paired, not that they exist in higher concentration. They may also exist in higher concentrarion. But, strictly speaking, that's not what the addition of an Al in that place means. I see. You have right - clustering effect can not be recalculated into doping level. Instead of that I should randomly distributed Al with the respect ration Al ions to all ions for desired concentration. OK. This is the follow-up to the point I made above. The convergence test on the host material sets an upper bound on how far away the second Al atom can be and still be visible in the calculation. You are now probing that with an actual placement of the Al atom. but why I do not observe any change in spectra? MFP is relevant, but it is just a way of putting a number to a more complicated effect. The bottom line is that you want to understand spectral trends as changes are made to the configuration used in a Feff claculation. To do that, you have to make a lot of Feff calculations. you mean thousands or hundreds thousands? ;)
Just, to follow up on Bruce's comment: A single cluster with 1 Al absorber and no Al scatterers does not really describe "1%" or even "0.5%" doping. It is "0%" doping. Dear Matt, I mixed up effect of clustering with doping, as Bruce explained. Anyway I can not agree with you - even one single Al is not 0% doping, almost 0 = yes...
By "1% Al doping", you (probably) really mean that each Zn site has a 1% chance of being occupied by Al. I think that's what you have to simulate. That means generated many (several hundred I would expect) clusters with each Zn site having 1% Al occupancy, and summing the results of those calculations. Unfortunately you have right - unfortunately because it needs many calculations...
Thanks for answers Darek
Hi Kicaj, Just, to follow up on Bruce's comment: A single cluster with 1 Al absorber and no Al scatterers does not really describe "1%" or even "0.5%" doping. It is "0%" doping. By "1% Al doping", you (probably) really mean that each Zn site has a 1% chance of being occupied by Al. I think that's what you have to simulate. That means generated many (several hundred I would expect) clusters with each Zn site having 1% Al occupancy, and summing the results of those calculations. I don't think the mean free path or size of the subcluster for the full multiple scattering matter here, except as it guides how large a cluster you need to consider. The size of the cluster needed for convergence is important, but does not really dictate an occupancy. What you want it a set of clusters that does simulate the proper doping level. --Matt
participants (3)
-
"Dr. Dariusz A. Zając" -
Bruce Ravel -
Matt Newville