Re: [Ifeffit] Differences between Larch and Artemis when performing FEFF calculations
Hello Matt! thank you for the response. In putting together all the project files for the minimal working example I tested a bunch more .cif files of various materials including different graphite file, and there was no difference in Larch and Athena in any of them (as long as the same FEFF version and input settings are used) as expected. Looking closely at the feff input file for the C.cif I was having trouble with, there were some differences which I haven't noticed when looking through them before. So as far as I am concerned, this was a user error on my part, as the generated input files weren't exactly the same. If I save the input feff file from Larch and use that in Artemis or the other way around, as you suggested, the results match exactly. The compared spectra in my previous message were calculated separately, exported and plotted together. thank you for very much for the help and kind regards, Ava Message: 1 Date: Mon, 16 Oct 2023 15:24:26 -0500 From: Matt Newville To: XAFS Analysis using Ifeffit Subject: Re: [Ifeffit] Differences between Larch and Artemis when performing FEFF calculations Message-ID: Content-Type: text/plain; charset="utf-8" Hi Ava, I think you'll have to give us more details about what you have done. The figures you show are all in R-space, so after at least some processing... So, yes, project files and/or scripts would be helpful. Yes, there can be subtle changes in the background subtraction (and in the normalization process too) between Larch and Ifeffit/Athena/Artemis. By default, Artemis uses Feff 6.10 and Larch uses Feff 8. For the C K edge, that could have a noticeable difference, especially in the placing of the k=0 value, though I do not know how big that effect would be for C (graphite?). But also, Artemis and Larch can both read the inputs from the other calculations: it might be that this is what you have done to make the plots, but that wasn't 100% clear to me. On Mon, Oct 16, 2023 at 2:08?PM Ava Rajh wrote: Dear all! I haven't been able to find a similar question/issue in the previous threads, so I hope someone can help me figure out what is going on. I am trying to fit a Carbon EXAFS spectra, using graphite as a model. I am first focusing on just the first two single scattering paths, so I calculated the theoretical paths with FEFF in Larch and tried using them to fit the spectra. I was consistently getting slightly lower distances than expected, but otherwise an OK fit. The issue is, I tried to compare the analysis with a colleague who is using Athena. At first glance the EXAFS spectra, using the exact same parameters, looked very similar (but not exactly the exactly the same, this I attributed to Larch using a different autobk procedure). I would have however expected the theoretical paths to match exactly, if they were calculated and plotted with the same parameters. But they were also slightly different. I then downloaded Athena and spent time trying to find where the differences come from. If I compare the first two calculated shells from Larch with the ones from Athena, with exactly the same set of test parameters (S02 = 1, E0 = 0, dr1 = 0, s2_1 = 0, dr2 = 0, s2_2 = 0), the resulting models do not match. I made sure the paths are calculated from the came .cif file in both cases, use FEFF6, have the same calculated reference distances, same FT... So, my main question is, am I missing something important in regards to calculations, why would the calculated paths be different and which one would be the "correct" one to use for the fit? And the other question would be about the fact that EXAFS spectra of experimental data look slightly different using Larch and Athena, am I right in disregarding this, or should I dig deeper and find the source of discrepancy? I am enclosing a plot of just the calculated first two shells from Athena and Larch (FT: kmin = 2, kmax = 7.5, Fittting in R space, kw = 3, kWindow = Hanning, dk = 1.0, Rmin = 0.6, Rmax = 2.1) along with the cif file I ended up using for testing the differences. If it would be helpful, I can also provide the project files and larch script I used for the dataset, but I am mainly interested in understanding the differences seen in the theoretical parts first. I tested this using Larch v 0.9.72 and Demeter 0.9.26 Thank you ver much for the help, and If I need to provide any additional info please let me know. kind regards, Ava -- Ava Rajh___ ___ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit
Re: [Ifeffit] Artemis fit error - chi-square and R-factors are always equal to 0
HI, I hope I am replying to the thread correctly. Thank you for your insights, I have spent a lot of time tinkering with installation and I have a partial solution. To first answer your questions: - yes, the fit ran ok and comparing it to the same fit using ifeffit (done by a a machine using ifeffit that works correctly), the calculated parameters are almost the same, with slight differences of 10-5 order of magnitude. The plot after the fit also looks fine. -the log I attached was from the example Cu data set, but the issue happens on every fit I try to run. -I'm not sure about the feff.dat file used, I haven't been able to locate it in my installation. - Feff6 (or Feff6l if I set it in preferences) are able to run as far as I can tell -I have run the example larch scripts to which you pointed me, and most of them were able to run ok, with no obvious issues (the parameters looked ok and reduced chi-square and R vales were reported). There was an error in running a few of the tests something to do with PlotDisplay, but I'm pretty sure these aren't related issues. To me, all of this points to some trouble Artemis has communicating with larch. I have tried downgrading the installation from 0.9.50 (that installs with pip) to 0.9.49 (that the website says is the current) and an even older ones but no luck. The solution war disappointingly trivial: If I completely removed larch and demeter suite along with everything else I installed in the last few days, from my computer and tried to compile it again, this time with ifeffit, I finally managed to get the reduced-chi square and R values. I have no idea why it didn't work when I tried it a few times before before but now I am able to get the fit and all required parameters. It still doesn't work with larch, but for now, at least I'm able to get the results. I'll re-install Larch and tinker with it a bit more to try to figure out where the error was coming from and report back if I find anything that might help someone else facing the same issue. Thanks again for the help Best, Ava ___ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit
[Ifeffit] Artemis fit error - chi-square and R-factors are always equal to 0
Hello, I have recently installed Demeter suite on my LinuxMint 20 machine with all required dependencies met. Athena works well and Demeter passed all tests during installation. When running Artemis however, there is an issue with fits that persists with provided examples any any other projects I open (including the ones done by colleagues and just re-run on my machine). After the fit (which completes fine and returns expected fit that looks ok when comparing it to experiment), when inspecting the log file, chi-square, reduced chi-square and R values are equal to 0 (example log file attached bellow). No errors accompany the fit. I am using Larch installed with anaconda, and Artemis uses Feff6 executable. I have re-installed Demeter and tried to get it to work with Ifeffit but the error persists. I have found an existing git hub thread dealing with similar issue (https://github.com/bruceravel/demeter/issues/62) but no solution that I could discern. Could it be an issue with my perl version (5.30.0)? If someone has any ideas what the issue could be and how to fix it, I would appreciate the help. Please let me know if I can provide any additional information, and I apologize if there is something obvious I may have missed. Sincerely, Ava -- Ava Rajh Name: Fit 4(fzexv) Description : fit to cu010k Figure of merit : 4 Time of fit : 2021-03-26T12:12:21 Environment : Demeter 0.9.26 with perl 5.03 and using Larch X.xx on linux Interface : Artemis (Wx 0.9932) Prepared by : ava@ava-PC Contact : =*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*= Independent points : 25.401 Number of variables : 4 Chi-square : 0.000 Reduced chi-square : 0.000 R-factor: 0.000 Number of data sets : 1 Happiness = 100.00/100 color = #D8E796 * Note: happiness is a semantic parameter and should * *NEVER be reported in a publication -- NEVER!* guess parameters: ss_Cu1 = 0.00341875# +/- 0.3724 [0.00300] dr_Cu1 = -0.00471435# +/- 0.00045738 [-0.00504] dE0= 5.41536257# +/- 0.14032983 [5.36177] amp= 0.89986572# +/- 0.00826589 [0.9] set parameters: N1 = 1. Correlations between variables: All other correlations below 0.4 = Data set >> cu010k << : file= : name= cu010k : k-range = 3.000 - 22.950 : dk = 1 : k-window= Hanning : k-weight= 3 : R-range = 1 - 3 : dR = 0.0 : R-window= Hanning : fitting space = r : background function = no : phase correction= no : background removal = : user-supplied epsilon_k = 0 : epsilon_k by k-weight = 3 -> 2.602e-04 : epsilon_r by k-weight = 3 -> 3.592e-01 : R-factor by k-weight= 1 -> 0.00434, 2 -> 0.00233, 3 -> 0.00246 nameN S02 sigma^2 e0 delr Reff R = [atoms] Cu.1 12.000 0.000 0.0 0.000 0.0 2.55270 2.55270 nameei third fourth = [atoms] Cu.1 0.0 0.0 0.0 =*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*=*= ___ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit Unsubscribe: http://millenia.cars.aps.anl.gov/mailman/options/ifeffit