October 2010 - Ifeffit - millenia.cars.aps.anl.gov

feff question
by George Sterbinsky 03 Nov '10

03 Nov '10

Hi, I am hoping someone can explain to me how to run Feff 7 on a mac (OSX version 10.5.8) starting from the feff702.f file. Thank you, George

3 10

"Hard tests failed in fovrg"
by srihari velaga 29 Oct '10

29 Oct '10

Hi all, I am trying to fit Cd kedge EXAFS data of CdO and CdO-CaO solid solution. Like many others I ended with “Hard tests failed in fovrg. Muffin-tin radius may be too large; coordination number too small“error. Professor John, mentioned this may be due to insufficient grid size in calculating the phase shifts and he advised to post the "feff.inp", to have a look at it. But I have 38 feff.inp files in total (corresponding to CdO, solid solutions of CdO-CaO at various compositions and for dopents as well). I feel it will be hard to look into this may files, for any one. I will be thankful if anyone help me, in how to overcome this problem. Here with I am attaching the "feff.inp" of CdO. Please tell me how to modify feff.inp file such that I can do myself for the rest of feff files Thanks in advance. Srihari. V VELAGA.SRIHARI, DAE RESEARCH FELLOW, X-RAY AND CRYSTAL GROWTH SECTION, MATERIAL SCIENCE DIVISION, INDIRA GANDHI CENTRE FOR ATOMIC RESEARCH, KALPAKKAM,TAMILNADU-603102, INDIA.

1 0

Re: [Ifeffit] Asymmetric error bars in IFeffit
by Scott Calvin 25 Oct '10

25 Oct '10

Yes; it's a case of trying to distinguish between a few boulders and lots of pebbles; the total volume isn't the issue. What I'm looking at is something like surface/volume ratio, but with "surface" being path-dependent and gradual. For a nearest-neighbor path, only the top monolayer of atoms are on the surface. For a 5 angstrom path, the transition region from "surface" to "core" extends 5 angstroms in. But that more sophisticated definition of "surface" doesn't change the fact that the dominant dependence is 1/R, so that should address the issue. --Scott Calvin Sarah Lawrence College On Oct 25, 2010, at 4:43 AM, Matt Newville wrote: > Hi Scott, > > That's a pretty amazing use case. > > But I'm not sure I understand the issue exactly right. I would have > thought the volume (r**3) was the important physical parameter, and > that a 1000nm particle would dominate the spectra over 3nm particles. > Or is it that you are trying to distinguish between 1 very large > crystal or 100s of smaller crystals? Perhaps the effect you're > really trying to account for is the surface/volume ratio? If so, I > think using Matthew Marcus's suggestion of using 1/r (with a safety > margin) makes the most sense. > > --Matt >

1 0

Re: [Ifeffit] LCF analysis
by Scott Calvin 25 Oct '10

25 Oct '10

On Oct 25, 2010, at 8:25 AM, Wayne W Lukens Jr wrote: > A more useful way to look at this is that the probabilities that A, > B and C are present are 99.99999999%, 93%, and 77%, respectively. An excellent post, Wayne, but I don't think that last statement is quite right. If the F-test gives a probability of 0.23 for material C, I believe it's saying that there is a 23% that, given the noise level in the data, the fit would indicate that C was present when it was not. That is not the same thing as saying there is a 77% chance of C being present. To see this, imagine very, very noisy data. Including C in the fit might very well improve the fit in the sense of an R-factor--maybe, in fact, there's a 45% chance of a modest improvement with a given set of very noisy data, even if there's no C present. That does not mean that a result like that should lead to the conclusion that C is more likely than not present (55%). --Scott Calvin Sarah Lawrence College

3 2

LCF analysis
by Luxton.Todd＠epamail.epa.gov 25 Oct '10

25 Oct '10

I have a general question for the group. How do you conduct your LCF fitting routine? Do most people allow Eo to vary for the standards, or force the weights of all of the combinations to 1? How do people judge a significant improvement to the resulting goodness of fit parameters? I am aware of most of the pitfalls associated with LCF analysis of XAFS data (over fitting data with too many standards, not having information to support the use of a standard in the fit, ect.). What I am really interested in is how people fit and judge the quality of a fit after collecting all of the necessary background information? Or, what is the process/routine people go through during LCF analysis? I have looked at older threads on the IFEFFIT web page, but I was unable to find a description of the general process people follow. Additionally, if anyone is aware of a reference discussing the potential problems and/or solutions to fitting data using LCF that would be great. All the Best Todd

2 1

Asymmetric error bars in IFeffit
by Scott Calvin 25 Oct '10

25 Oct '10

Hi all, I'm puzzling over an issue with my latest analysis, and it seemed like the sort of thing where this mailing list might have some good ideas. First, a little background on the analysis. It is a simultaneous fit to four samples, made of various combinations of three phases. Mossbauer has established which samples include which phases. One of the phases itself has two crystallographically inequivalent absorbing sites. The result is that the fit includes 12 Feff calculations, four data sets, and 1000 paths. Remarkably, everything works quite well, yielding a satisfying and informative fit. Depending on the details, the fit takes about 90 minutes to run. Kudos to Ifeffit and Horae for making such a thing possible! Several of the parameters that the fit finds are "characteristic crystallite radii" for the individual phases. In my published fits, I often include a factor that accounts for the fact that a phase is nanoscale in a crude way: it assumes the phase is present as spheres of uniform radius and applies a suppression factor to the coordination numbers of the paths as a function of that radius and of the absorber- scatterer distance. Even though this model is rarely strictly correct in terms of morphology and size dispersion, it gives a first-order approximation to the effect of the reduced coordination numbers found in nanoscale materials. Some people, notably Anatoly Frenkel, have published models which deal with this effect much more realistically. But those techniques also require more fitted variables and work best with fairly well-behaved samples. I tend to work with "messy" chemical samples of free nanoparticles where the assumption of sphericity isn't terrible, and the size dispersion is difficult to model accurately. At any rate, the project I'm currently working on includes a fitted characteristic radius of the type I've described for each of the phases in each of the samples. And again, it seems to work pretty well, yielding values that are plausible and largely stable. That's the background information. Now for my question: The effect of the characteristic radius on the spectrum is a strongly nonlinear function of that radius. For example, the difference between the EXAFS spectra of 100 nm and 1000 nm single crystals due to the coordination number effect is completely negligible. The difference between 1 nm and 10 nm crystals, however, is huge. So for very small crystallites, IFeffit reports perfectly reasonable error bars: the radius is 0.7 +/- 0.3 nm, for instance. For somewhat larger crystallites, however, it tends to report values like 10 +/- 500 nm. I understand why it does that: it's evaluating how much the parameter would have to change by to have a given impact on the chi square of the fit. And it turns out that once you get to about 10 nm, the size could go arbitrarily higher than that and not change the spectrum much at all. But it couldn't go that much lower without affecting the spectrum. So what IFeffit means is something like "the best fit value is 10 nm, and it is probable that the value is at least 4 nm." But it's operating under the assumption that the dependence of chi-square on the parameter is parabolic, so it comes up with a compromise between a 6 nm error bar on the low side and an infinitely large error bar on the high side. Compromising with infinity, however, rarely yields sensible results. Thus my question is if anyone can think of a way to extract some sense of these asymmetric error bars from IFeffit. Here are possibilities I've considered: --Fit something like the log of the characteristic radius, rather than the radius itself. That creates an asymmetric error bar for the radius, but the asymmetry the new error bar possesses has no relationship to the uncertainty it "should" possess. This seems to me like it's just a way of sweeping the problem under the rug and is potentially misleading. --Rerun the fits setting the variable in question to different values to probe how far up or down it can go and have the same effect on the fit. But since I've got nine of these factors, and each fit takes more than an hour, the computer time required seems prohibitive! --Somehow parameterize the guessed variable so that it does tend to have symmetric error bars, and then calculate the characteristic radius and its error bars from that. But it's not at all clear what that parameterization would be. --Ask the IFeffit mailing list for ideas! Thanks! --Scott Calvin Sarah Lawrence College

3 2

Re: [Ifeffit] Asymmetric error bars in IFeffit
by Scott Calvin 25 Oct '10

25 Oct '10

I don't think it's at all strange, Anatoly, and I think Matthew's solution is the right one--it seems obvious in retrospect that the parameter that Ifeffit should evaluate is 1/R, but apparently it wasn't obvious to me on Friday. :) As for obtaining N instead of R, the beauty of both of our algorithms is that they don't depend on finding N; they depend on finding the ratio of N's for different shells. Finding N accurately is notoriously challenging: you need some way of getting S02, you need to have the normalization right, and you're sunk if there are data quality issues like an inhomogeneous sample, uncorrected self-absorption, or significant beam harmonics. But finding the ratio of N for two or more different shells doesn't depend so strongly on any of those things. Since my method implicitly involves multiple ratios of coordination numbers, it is not so clear how to invert it. In any case, I expect Matthew's solution to work, and will pursue it further on Monday. --Scott Calvin Sarah Lawrence College On Oct 24, 2010, at 5:59 PM, Frenkel, Anatoly wrote: > Scott, > It is a strange result. Suppose you fit a bulk metal foil and vary > the 1nn coordination number. You will not get 12 +/- 1000. You will > get about 12 +/- 0.3 depending on the data quality and the k range, > and on the amplitude factor you fix constant. Then, suppose you take > your formula for a particle radius from your JAP article and > propagate this uncertainty to get the radius uncertainty. That would > give you a huge error because you are in the flat region of the N(R) > function and R does bit affect N. > The meaning of your large error bar is, I think, that you are in > such a large limit of sizes that they cannot be inverted to get N > and thus the errors cannot be propagated to find Delta R. > Why don't you try to obtain N instead of R? You will get much > smaller error bars and you can find the lower R limit from your N(R) > equation (by plugging in N - deltaN you will find R - delta R). > > The right limit is infinity as you pointed out. > > Anatoly > > ________________________________ > From: ifeffit-bounces(a)millenia.cars.aps.anl.gov <ifeffit-bounces(a)millenia.cars.aps.anl.gov > > > To: XAFS Analysis using Ifeffit <ifeffit(a)millenia.cars.aps.anl.gov> > Sent: Fri Oct 22 16:23:08 2010 > Subject: [Ifeffit] Asymmetric error bars in IFeffit > > Hi all, > > I'm puzzling over an issue with my latest analysis, and it seemed > like the sort of thing where this mailing list might have some good > ideas. > > First, a little background on the analysis. It is a simultaneous fit > to four samples, made of various combinations of three phases. > Mossbauer has established which samples include which phases. One of > the phases itself has two crystallographically inequivalent > absorbing sites. The result is that the fit includes 12 Feff > calculations, four data sets, and 1000 paths. Remarkably, everything > works quite well, yielding a satisfying and informative fit. > Depending on the details, the fit takes about 90 minutes to run. > Kudos to Ifeffit and Horae for making such a thing possible! > > Several of the parameters that the fit finds are "characteristic > crystallite radii" for the individual phases. In my published fits, > I often include a factor that accounts for the fact that a phase is > nanoscale in a crude way: it assumes the phase is present as spheres > of uniform radius and applies a suppression factor to the > coordination numbers of the paths as a function of that radius and > of the absorber-scatterer distance. Even though this model is rarely > strictly correct in terms of morphology and size dispersion, it > gives a first-order approximation to the effect of the reduced > coordination numbers found in nanoscale materials. Some people, > notably Anatoly Frenkel, have published models which deal with this > effect much more realistically. But those techniques also require > more fitted variables and work best with fairly well-behaved > samples. I tend to work with "messy" chemical samples of free > nanoparticles where the assumption of sphericity isn't terrible, and > the size dispersion is difficult to model accurately. > > At any rate, the project I'm currently working on includes a fitted > characteristic radius of the type I've described for each of the > phases in each of the samples. And again, it seems to work pretty > well, yielding values that are plausible and largely stable. > > That's the background information. Now for my question: > > The effect of the characteristic radius on the spectrum is a > strongly nonlinear function of that radius. For example, the > difference between the EXAFS spectra of 100 nm and 1000 nm single > crystals due to the coordination number effect is completely > negligible. The difference between 1 nm and 10 nm crystals, however, > is huge. > > So for very small crystallites, IFeffit reports perfectly reasonable > error bars: the radius is 0.7 +/- 0.3 nm, for instance. For somewhat > larger crystallites, however, it tends to report values like 10 +/- > 500 nm. I understand why it does that: it's evaluating how much the > parameter would have to change by to have a given impact on the chi > square of the fit. And it turns out that once you get to about 10 > nm, the size could go arbitrarily higher than that and not change > the spectrum much at all. But it couldn't go that much lower without > affecting the spectrum. So what IFeffit means is something like "the > best fit value is 10 nm, and it is probable that the value is at > least 4 nm." But it's operating under the assumption that the > dependence of chi-square on the parameter is parabolic, so it comes > up with a compromise between a 6 nm error bar on the low side and an > infinitely large error bar on the high side. Compromising with > infinity, however, rarely yields sensible results. > > Thus my question is if anyone can think of a way to extract some > sense of these asymmetric error bars from IFeffit. Here are > possibilities I've considered: > > --Fit something like the log of the characteristic radius, rather > than the radius itself. That creates an asymmetric error bar for the > radius, but the asymmetry the new error bar possesses has no > relationship to the uncertainty it "should" possess. This seems to > me like it's just a way of sweeping the problem under the rug and is > potentially misleading. > > --Rerun the fits setting the variable in question to different > values to probe how far up or down it can go and have the same > effect on the fit. But since I've got nine of these factors, and > each fit takes more than an hour, the computer time required seems > prohibitive! > > --Somehow parameterize the guessed variable so that it does tend to > have symmetric error bars, and then calculate the characteristic > radius and its error bars from that. But it's not at all clear what > that parameterization would be. > > --Ask the IFeffit mailing list for ideas! > > Thanks! > > --Scott Calvin > Sarah Lawrence College > <ATT00001..txt>

1 0

Re: [Ifeffit] Asymmetric error bars in IFeffit
by Frenkel, Anatoly 25 Oct '10

25 Oct '10

Scott, It is a strange result. Suppose you fit a bulk metal foil and vary the 1nn coordination number. You will not get 12 +/- 1000. You will get about 12 +/- 0.3 depending on the data quality and the k range, and on the amplitude factor you fix constant. Then, suppose you take your formula for a particle radius from your JAP article and propagate this uncertainty to get the radius uncertainty. That would give you a huge error because you are in the flat region of the N(R) function and R does bit affect N. The meaning of your large error bar is, I think, that you are in such a large limit of sizes that they cannot be inverted to get N and thus the errors cannot be propagated to find Delta R. Why don't you try to obtain N instead of R? You will get much smaller error bars and you can find the lower R limit from your N(R) equation (by plugging in N - deltaN you will find R - delta R). The right limit is infinity as you pointed out. Anatoly ________________________________ From: ifeffit-bounces(a)millenia.cars.aps.anl.gov <ifeffit-bounces(a)millenia.cars.aps.anl.gov> To: XAFS Analysis using Ifeffit <ifeffit(a)millenia.cars.aps.anl.gov> Sent: Fri Oct 22 16:23:08 2010 Subject: [Ifeffit] Asymmetric error bars in IFeffit Hi all, I'm puzzling over an issue with my latest analysis, and it seemed like the sort of thing where this mailing list might have some good ideas. First, a little background on the analysis. It is a simultaneous fit to four samples, made of various combinations of three phases. Mossbauer has established which samples include which phases. One of the phases itself has two crystallographically inequivalent absorbing sites. The result is that the fit includes 12 Feff calculations, four data sets, and 1000 paths. Remarkably, everything works quite well, yielding a satisfying and informative fit. Depending on the details, the fit takes about 90 minutes to run. Kudos to Ifeffit and Horae for making such a thing possible! Several of the parameters that the fit finds are "characteristic crystallite radii" for the individual phases. In my published fits, I often include a factor that accounts for the fact that a phase is nanoscale in a crude way: it assumes the phase is present as spheres of uniform radius and applies a suppression factor to the coordination numbers of the paths as a function of that radius and of the absorber-scatterer distance. Even though this model is rarely strictly correct in terms of morphology and size dispersion, it gives a first-order approximation to the effect of the reduced coordination numbers found in nanoscale materials. Some people, notably Anatoly Frenkel, have published models which deal with this effect much more realistically. But those techniques also require more fitted variables and work best with fairly well-behaved samples. I tend to work with "messy" chemical samples of free nanoparticles where the assumption of sphericity isn't terrible, and the size dispersion is difficult to model accurately. At any rate, the project I'm currently working on includes a fitted characteristic radius of the type I've described for each of the phases in each of the samples. And again, it seems to work pretty well, yielding values that are plausible and largely stable. That's the background information. Now for my question: The effect of the characteristic radius on the spectrum is a strongly nonlinear function of that radius. For example, the difference between the EXAFS spectra of 100 nm and 1000 nm single crystals due to the coordination number effect is completely negligible. The difference between 1 nm and 10 nm crystals, however, is huge. So for very small crystallites, IFeffit reports perfectly reasonable error bars: the radius is 0.7 +/- 0.3 nm, for instance. For somewhat larger crystallites, however, it tends to report values like 10 +/- 500 nm. I understand why it does that: it's evaluating how much the parameter would have to change by to have a given impact on the chi square of the fit. And it turns out that once you get to about 10 nm, the size could go arbitrarily higher than that and not change the spectrum much at all. But it couldn't go that much lower without affecting the spectrum. So what IFeffit means is something like "the best fit value is 10 nm, and it is probable that the value is at least 4 nm." But it's operating under the assumption that the dependence of chi-square on the parameter is parabolic, so it comes up with a compromise between a 6 nm error bar on the low side and an infinitely large error bar on the high side. Compromising with infinity, however, rarely yields sensible results. Thus my question is if anyone can think of a way to extract some sense of these asymmetric error bars from IFeffit. Here are possibilities I've considered: --Fit something like the log of the characteristic radius, rather than the radius itself. That creates an asymmetric error bar for the radius, but the asymmetry the new error bar possesses has no relationship to the uncertainty it "should" possess. This seems to me like it's just a way of sweeping the problem under the rug and is potentially misleading. --Rerun the fits setting the variable in question to different values to probe how far up or down it can go and have the same effect on the fit. But since I've got nine of these factors, and each fit takes more than an hour, the computer time required seems prohibitive! --Somehow parameterize the guessed variable so that it does tend to have symmetric error bars, and then calculate the characteristic radius and its error bars from that. But it's not at all clear what that parameterization would be. --Ask the IFeffit mailing list for ideas! Thanks! --Scott Calvin Sarah Lawrence College

1 0

Re: [Ifeffit] Asymmetric error bars in IFeffit
by Frenkel, Anatoly 23 Oct '10

23 Oct '10

Matthew, if you relied on FEFF as one part of the complex calculation, where the other part was the experimentally extracted one, you could've as well done everything with just FEFF. Why didn't you? A. ________________________________ From: ifeffit-bounces(a)millenia.cars.aps.anl.gov <ifeffit-bounces(a)millenia.cars.aps.anl.gov> To: XAFS Analysis using Ifeffit <ifeffit(a)millenia.cars.aps.anl.gov> Sent: Fri Oct 22 18:01:17 2010 Subject: Re: [Ifeffit] Asymmetric error bars in IFeffit Another reason, from my point of view, is that FEFF wasn't accurate enough to use on its own without references. Also, it still can be argued that the process of data reduction and filtering produces distortions which aren't captured by using FEFF alone. Further, if one is comparing very similar systems, e.g. bulk and nano of the same stuff, then with the exception of multiple scattering, the one system should be an ideal reference for the other. A trick I used to do: Suppose, for instance, that I wanted to fit a shell of Zn surrounded by Si (not an actual case). There's no Si-rich zinc silicide to use as a reference, but it's not too hard to make theta-CuAl2, a compound in which all Cu atoms are surrounded by Al in the first shell (this was done). From this, amp and phase functions could be extracted which refer to Cu looking at Al (Cu->Al). Next, to transform this into the desired Zn->Si, I would do FEFF calculations on identical structures with the atoms changed around and take: A(Zn->Si, semi-empirical) = A(Cu->Al, expt)*[A(Zn->Si, FEFF)/A(Cu->Al, FEFF)] phi(Zn->Si, semi-empirical) = phi(Cu->Al, expt)+[Phi(Zn->Si, FEFF)-phi(Cu->Al, FEFF)] with A, phi being amplitude and phase for a given shell, and appropriate account being taken of the distance differences involved. This makes sense if you consider A~ = A*exp(i*phi) to be one of the factors in a complex chi~ such that chi = Im(chi~), and you're essentially making a correction to ln(A~). Yes, this was low-rent and subject to errors, but it seemed to make sense provided one didn't try to take it too far, for instance trying to change Al for Au or an oxide for a metal. Brings back memories, not all of them fond :-) mam ----- Original Message ----- From: Frenkel, Anatoly <mailto:frenkel@bnl.gov> To: ifeffit(a)millenia.cars.aps.anl.gov Sent: Friday, October 22, 2010 2:38 PM Subject: Re: [Ifeffit] Asymmetric error bars in IFeffit On a related subject, now I understand why we use the concept of chemical transferability of amplitudes and phases by recycling the same FEFF path for different systems. The true reason is historic: back then it took one hour for one FEFF calculation.... Anatoly ________________________________ From: ifeffit-bounces(a)millenia.cars.aps.anl.gov <ifeffit-bounces(a)millenia.cars.aps.anl.gov> To: XAFS Analysis using Ifeffit <ifeffit(a)millenia.cars.aps.anl.gov> Sent: Fri Oct 22 16:23:08 2010 Subject: [Ifeffit] Asymmetric error bars in IFeffit Hi all, I'm puzzling over an issue with my latest analysis, and it seemed like the sort of thing where this mailing list might have some good ideas. First, a little background on the analysis. It is a simultaneous fit to four samples, made of various combinations of three phases. Mossbauer has established which samples include which phases. One of the phases itself has two crystallographically inequivalent absorbing sites. The result is that the fit includes 12 Feff calculations, four data sets, and 1000 paths. Remarkably, everything works quite well, yielding a satisfying and informative fit. Depending on the details, the fit takes about 90 minutes to run. Kudos to Ifeffit and Horae for making such a thing possible! Several of the parameters that the fit finds are "characteristic crystallite radii" for the individual phases. In my published fits, I often include a factor that accounts for the fact that a phase is nanoscale in a crude way: it assumes the phase is present as spheres of uniform radius and applies a suppression factor to the coordination numbers of the paths as a function of that radius and of the absorber-scatterer distance. Even though this model is rarely strictly correct in terms of morphology and size dispersion, it gives a first-order approximation to the effect of the reduced coordination numbers found in nanoscale materials. Some people, notably Anatoly Frenkel, have published models which deal with this effect much more realistically. But those techniques also require more fitted variables and work best with fairly well-behaved samples. I tend to work with "messy" chemical samples of free nanoparticles where the assumption of sphericity isn't terrible, and the size dispersion is difficult to model accurately. At any rate, the project I'm currently working on includes a fitted characteristic radius of the type I've described for each of the phases in each of the samples. And again, it seems to work pretty well, yielding values that are plausible and largely stable. That's the background information. Now for my question: The effect of the characteristic radius on the spectrum is a strongly nonlinear function of that radius. For example, the difference between the EXAFS spectra of 100 nm and 1000 nm single crystals due to the coordination number effect is completely negligible. The difference between 1 nm and 10 nm crystals, however, is huge. So for very small crystallites, IFeffit reports perfectly reasonable error bars: the radius is 0.7 +/- 0.3 nm, for instance. For somewhat larger crystallites, however, it tends to report values like 10 +/- 500 nm. I understand why it does that: it's evaluating how much the parameter would have to change by to have a given impact on the chi square of the fit. And it turns out that once you get to about 10 nm, the size could go arbitrarily higher than that and not change the spectrum much at all. But it couldn't go that much lower without affecting the spectrum. So what IFeffit means is something like "the best fit value is 10 nm, and it is probable that the value is at least 4 nm." But it's operating under the assumption that the dependence of chi-square on the parameter is parabolic, so it comes up with a compromise between a 6 nm error bar on the low side and an infinitely large error bar on the high side. Compromising with infinity, however, rarely yields sensible results. Thus my question is if anyone can think of a way to extract some sense of these asymmetric error bars from IFeffit. Here are possibilities I've considered: --Fit something like the log of the characteristic radius, rather than the radius itself. That creates an asymmetric error bar for the radius, but the asymmetry the new error bar possesses has no relationship to the uncertainty it "should" possess. This seems to me like it's just a way of sweeping the problem under the rug and is potentially misleading. --Rerun the fits setting the variable in question to different values to probe how far up or down it can go and have the same effect on the fit. But since I've got nine of these factors, and each fit takes more than an hour, the computer time required seems prohibitive! --Somehow parameterize the guessed variable so that it does tend to have symmetric error bars, and then calculate the characteristic radius and its error bars from that. But it's not at all clear what that parameterization would be. --Ask the IFeffit mailing list for ideas! Thanks! --Scott Calvin Sarah Lawrence College ________________________________ _______________________________________________ Ifeffit mailing list Ifeffit(a)millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit

2 1

Re: [Ifeffit] Asymmetric error bars in IFeffit
by Frenkel, Anatoly 22 Oct '10

22 Oct '10

On a related subject, now I understand why we use the concept of chemical transferability of amplitudes and phases by recycling the same FEFF path for different systems. The true reason is historic: back then it took one hour for one FEFF calculation.... Anatoly ________________________________ From: ifeffit-bounces(a)millenia.cars.aps.anl.gov <ifeffit-bounces(a)millenia.cars.aps.anl.gov> To: XAFS Analysis using Ifeffit <ifeffit(a)millenia.cars.aps.anl.gov> Sent: Fri Oct 22 16:23:08 2010 Subject: [Ifeffit] Asymmetric error bars in IFeffit Hi all, I'm puzzling over an issue with my latest analysis, and it seemed like the sort of thing where this mailing list might have some good ideas. First, a little background on the analysis. It is a simultaneous fit to four samples, made of various combinations of three phases. Mossbauer has established which samples include which phases. One of the phases itself has two crystallographically inequivalent absorbing sites. The result is that the fit includes 12 Feff calculations, four data sets, and 1000 paths. Remarkably, everything works quite well, yielding a satisfying and informative fit. Depending on the details, the fit takes about 90 minutes to run. Kudos to Ifeffit and Horae for making such a thing possible! Several of the parameters that the fit finds are "characteristic crystallite radii" for the individual phases. In my published fits, I often include a factor that accounts for the fact that a phase is nanoscale in a crude way: it assumes the phase is present as spheres of uniform radius and applies a suppression factor to the coordination numbers of the paths as a function of that radius and of the absorber-scatterer distance. Even though this model is rarely strictly correct in terms of morphology and size dispersion, it gives a first-order approximation to the effect of the reduced coordination numbers found in nanoscale materials. Some people, notably Anatoly Frenkel, have published models which deal with this effect much more realistically. But those techniques also require more fitted variables and work best with fairly well-behaved samples. I tend to work with "messy" chemical samples of free nanoparticles where the assumption of sphericity isn't terrible, and the size dispersion is difficult to model accurately. At any rate, the project I'm currently working on includes a fitted characteristic radius of the type I've described for each of the phases in each of the samples. And again, it seems to work pretty well, yielding values that are plausible and largely stable. That's the background information. Now for my question: The effect of the characteristic radius on the spectrum is a strongly nonlinear function of that radius. For example, the difference between the EXAFS spectra of 100 nm and 1000 nm single crystals due to the coordination number effect is completely negligible. The difference between 1 nm and 10 nm crystals, however, is huge. So for very small crystallites, IFeffit reports perfectly reasonable error bars: the radius is 0.7 +/- 0.3 nm, for instance. For somewhat larger crystallites, however, it tends to report values like 10 +/- 500 nm. I understand why it does that: it's evaluating how much the parameter would have to change by to have a given impact on the chi square of the fit. And it turns out that once you get to about 10 nm, the size could go arbitrarily higher than that and not change the spectrum much at all. But it couldn't go that much lower without affecting the spectrum. So what IFeffit means is something like "the best fit value is 10 nm, and it is probable that the value is at least 4 nm." But it's operating under the assumption that the dependence of chi-square on the parameter is parabolic, so it comes up with a compromise between a 6 nm error bar on the low side and an infinitely large error bar on the high side. Compromising with infinity, however, rarely yields sensible results. Thus my question is if anyone can think of a way to extract some sense of these asymmetric error bars from IFeffit. Here are possibilities I've considered: --Fit something like the log of the characteristic radius, rather than the radius itself. That creates an asymmetric error bar for the radius, but the asymmetry the new error bar possesses has no relationship to the uncertainty it "should" possess. This seems to me like it's just a way of sweeping the problem under the rug and is potentially misleading. --Rerun the fits setting the variable in question to different values to probe how far up or down it can go and have the same effect on the fit. But since I've got nine of these factors, and each fit takes more than an hour, the computer time required seems prohibitive! --Somehow parameterize the guessed variable so that it does tend to have symmetric error bars, and then calculate the characteristic radius and its error bars from that. But it's not at all clear what that parameterization would be. --Ask the IFeffit mailing list for ideas! Thanks! --Scott Calvin Sarah Lawrence College

2 1