January 2004 - Ifeffit - millenia.cars.aps.anl.gov

fitting with lots of variables
by k-kupiecki＠northwestern.edu 21 Jan '04

21 Jan '04

Hi, Just a quick question... Currently I'm fitting experimental spectra to feff results from two different atomic configurations to try and figure out which scenerio is a better fit and what the ultimate atom positions are.... In doing this I'm already fitting for different sigma^2 for each path (since I'm not dealing with a monoatomic system).... Now I think I would like to fit for each path's individual DelR (....previously I'd been fitting with eta*reff for each but I'm expecting the displacement about the core atom to be quite different than further away from it).... My problem is... I don't have enough independent points to accomodate all the variables given the paths included when fitting over a specific R range. Currently I figure I have at least 2 options: 1)(Assuming I'm already using the largest suitable k-range) -- Increase the r-range for the fit (i.e. if fitting the single scattering paths to 4 Ang requires 18 variables but I only have 15 independent points with r-range 1-4... perhaps increase the r-range 1-4.5 (18 ind. points in this case) or 1-5 (20 ind. points in this case) while still including the paths less than 4Ang in the fit. 2) Lump some of the paths' DelRs into expressions (i.e. having DelRs = eta1*reff, or eta2*reff, or eta3*reff etc for groupings of paths dependent upon distance from core atom). Any suggestions about these ideas, or any other ideas-- I greatly appreciate the input! Thanks a ton! -Kristine

3 2

Debye-Waller factors for MS paths
by Peter Southon 14 Jan '04

14 Jan '04

Dear everyone, One of the features of constructing models with Ifeffit is that we are forced to think long and hard about the parameters to use for each path. In particular I have had a great deal of difficulty in deciding how to approach the Debye-Waller factors for with multiple scattering (MS) paths. The ideal approach is to define the D-W factors for the MS paths in terms of the factors for the single scattering paths. That way you add no further variables to the model. But if you think about it you will realise that you cannot simply add all the relevant SS factors up to get a sum for your MS path - it all depends on the geometry, the direction of vibration, correlation of vibrations, etc. Its all a bit tricky for those of us dealing with intense MS paths. So what I am asking is, does anyone know of a general method or approach to define the Debye-Waller factors for MS paths (in terms of SS paths or otherwise)? How do all you Ifeffiters deal with the problem? I've found a few papers that discuss it (listed below) but they generally assume that all vibrations are uncorrelated, which is seldom a realistic situation in continuous solids. Munoz-Paez, A. (2000). Inorganic Chemistry 39(17): 3784-3790. Sakane, H. (1998). Journal of the American Chemical Society 120(40): 10397-10401. Haskel, D. (1998). PhD thesis, University of Washington. (appendix) http://www.aps.anl.gov/xfd/people/haskel/PS/thesis.ps Is there anything out there that I've missed? I'd be particularly interested in anything relating to continuous solids or correlated vibrations. Comments/questions/pearls of wisdom very welcome. Peter Peter Southon Research Fellow - School of Chemistry University of Sydney, NSW 2006, Australia +61 2 9351 4425

4 3

Re: Ifeffit Digest, Vol 11, Issue 5
by Grant Bunker 14 Jan '04

14 Jan '04

Hello, folks - In regard to Peter's question regarding multiple scattering DWFs: Matt was correct that there have been two approaches (to my knowledge) that attempt to directly address the problem: Poiarkova and Rehr's approach, and our approach. There are pros and cons to each. We (Nick Dimakis and I) have focussed on coordination complexes of metals with amino acid and similar ligands to directly address metalloproteins. The approach probably would work on other systems as well. Our method works like this: we calculate using DFT the normal modes (vibrational eigenvectors and eigenvaluesof model complexes as a function of interatomic distance and bond angles (as needed). We started this about ten years ago, and initially tried force constant models and various approximate quantum mechanical methods but found them inadequate. DFT level give sufficient accuracy and speed. Fortunately computers are fast and cheap nowadays. We have written a programs that take the eigenvalues and eigenvectors and the feff output and generates MS DWFs for each path. We have found that a "group fitting" approach works well, i.e. neglecting the inter-group multiple scattering to a first approximation. It has worked well in the cases we've tried. Since it isn't really practical to expect people to do these DFT calculations themselves (it's difficult to get convergence in many cases, it's easy to get trapped in a local minimum, and the minimization algorithms in the commercial codes are fairly antiquated), we have tried and succeeded in parameterizing the DWF for each MS path as a function of distance (and temperature). In almost all cases they are easily parameterized with polynomials as a function of both distance and temperature. So far we have done this with Zn coordinated to cysteine, histidine, and carboxylate ligands (e.g. glu), and also Cu(I) and Cu(II). We are working on Fe, Mn and other metal-ligand complexes as well, depending on funding. Hope that helps. Grant On Wed, 14 Jan 2004 ifeffit-request(a)millenia.cars.aps.anl.gov wrote: > Send Ifeffit mailing list submissions to > ifeffit(a)millenia.cars.aps.anl.gov > > To subscribe or unsubscribe via the World Wide Web, visit > http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit > or, via email, send a message with subject or body 'help' to > ifeffit-request(a)millenia.cars.aps.anl.gov > > You can reach the person managing the list at > ifeffit-owner(a)millenia.cars.aps.anl.gov > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of Ifeffit digest..." > > > Today's Topics: > > 1. Spline clamps (SCalvin100(a)aol.com) > 2. Re: Spline clamps (Bruce Ravel) > 3. Re: Spline clamps (Matt Newville) > 4. Debye-Waller factors for MS paths (Peter Southon) > 5. RE: Debye-Waller factors for MS paths (Kelly, Shelly D.) > 6. Re: Debye-Waller factors for MS paths (Bruce Ravel) > 7. Re: Debye-Waller factors for MS paths (Matt Newville) > 8. linux: wouldn't it be cool... (Bruce Ravel) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Tue, 13 Jan 2004 15:41:47 EST > From: SCalvin100(a)aol.com > Subject: [Ifeffit] Spline clamps > To: ifeffit(a)millenia.cars.aps.anl.gov > Message-ID: <1cf.17c88a73.2d35b20b(a)aol.com> > Content-Type: text/plain; charset="us-ascii" > > Hi all, > > Happy New Year to all! I've been busy pushing the new versions of Artemis and > Ifeffit pretty hard (3-edge fits, over 100 paths, also over 100 guess,def,set > parameters) and they seem to be behaving pretty nicely. There are certainly > some advantages and some disadvantages to this compared to the old FEFFIT > package; one major advantage is this will certainly reduced the learning curve for > new undergraduates in my group. > > I am posting now because I am not sure I understand the behavior of the > spline clamps in Ifeffit. When I set a top end to the spline range that is below > the top of the data range, the high spline clamp has very little, if any, > effect. (As a simple example, if I take the "clamp.prj" project that comes with > Athena and set all of the spline ranges to top out at 400 eV, all four backgrounds > look the same.) Is this what should be happening, or is this a bug? > > In case this is a bug, here's the version information: > > Athena 0.8.024 and ifeffit 1.2.4 running on Windows XP. > > I'll also be glad to send a project file to Bruce or Matt if they wish, > although it would just be the clamp.prj file with the top of the spline range > changed... > > --Scott >

1 0

linux: wouldn't it be cool...
by Bruce Ravel 14 Jan '04

14 Jan '04

Hi, I was just thinking that it would be nice to be able to install ifeffit and friends into either the KDE or GNOME menus. For example, one could imagine installing athena and artemis by doing perl Makefile.PL make make install make install-kde or make install-gnome where the last step makes the menu entries for the codes. Do any of the linux hacks out there know how this is done (at the level of a script -- I know how to do it by hand)? B -- Bruce Ravel ----------------------------------- ravel(a)phys.washington.edu Code 6134, Building 3, Room 222 Naval Research Laboratory phone: (1) 202 767 5947 Washington DC 20375, USA fax: (1) 202 767 1697 NRL Synchrotron Radiation Consortium (NRL-SRC) Beamlines X11a, X11b, X23b National Synchrotron Light Source Brookhaven National Laboratory, Upton, NY 11973 My homepage: http://feff.phys.washington.edu/~ravel EXAFS software: http://feff.phys.washington.edu/~ravel/software/exafs/

2 1

RE: [Ifeffit] Debye-Waller factors for MS paths
by Kelly, Shelly D. 14 Jan '04

14 Jan '04

Hi Peter, Please see below: > > Dear everyone, > > One of the features of constructing models with Ifeffit is > that we are > forced to think long and hard about the parameters to use for > each path. In > particular I have had a great deal of difficulty in deciding how to > approach the Debye-Waller factors for with multiple > scattering (MS) paths. > > The ideal approach is to define the D-W factors for the MS > paths in terms > of the factors for the single scattering paths. That way you > add no further > variables to the model. But if you think about it you will > realise that you > cannot simply add all the relevant SS factors up to get a sum > for your MS > path - it all depends on the geometry, the direction of vibration, > correlation of vibrations, etc. Its all a bit tricky for those of us > dealing with intense MS paths. > > So what I am asking is, does anyone know of a general method > or approach to > define the Debye-Waller factors for MS paths (in terms of SS paths or > otherwise)? How do all you Ifeffiters deal with the problem? > I've found a > few papers that discuss it (listed below) but they generally > assume that > all vibrations are uncorrelated, which is seldom a realistic > situation in > continuous solids. > > Munoz-Paez, A. (2000). Inorganic Chemistry 39(17): 3784-3790. > Sakane, H. (1998). Journal of the American Chemical Society 120(40): > 10397-10401. > Haskel, D. (1998). PhD thesis, University of Washington. (appendix) > http://www.aps.anl.gov/xfd/people/haskel/PS/thesis.ps > > Is there anything out there that I've missed? I'd be particularly > interested in anything relating to continuous solids or correlated > vibrations. Comments/questions/pearls of wisdom very welcome. > > Peter > > > Peter Southon > Research Fellow - School of Chemistry > University of Sydney, NSW 2006, Australia > +61 2 9351 4425 > There are a couple of ideas that I have found that work pretty well for determining sigma2 values for a lot of scattering paths. Idea 1: Try to use a correlated debye model or the eins model. You can have more than one debye/eins temperature depending on the atom types and path lengths. Idea 2: Group paths that are similar in distance and atom types and set their sigma2 values to the same value, including the M.S. paths. -use the abs function in defining a value for sigma2 to force it to be positive. sigma2 path1 abs(sig) guess sig 0.010 Idea 3: Usually, the triangle paths that have a really large degeneracy end up with their own sigma2 value. Most models can be close to working without these paths, and then added them at the end to clean up the fit. Idea 4: Linear paths can be quantified in terms of the sigma2 values of the single scatting paths that contain the end atoms. Another paper that talks about this is: E A Hudson, J J Rehr and J J Bucher. "Multiple-scattering calculations of the uranium L3-edge X-ray-absorption near-edge structure." Phys. Rev. B 52(19): pp 13815-3826, 1995. -For a forward linear path A-B-C, sigma2(A-B-C-A)=sigma2(A-B-C-B-A)=sigma2(A-C-A) -For a backward scattering linear path B1-A-B2, and if A is a lot heavier than B then Sigma2(A-B1-A-B2-A)~sigma2(A-B1-B2-A)~2xsigma2(A-B), and sigma2(A-B1-A-B1-A)=4xsigma2(A-B) (John Rehr and I had a previous discussion about this with the subject "relation between DW factors in uranyls") Golden Rule: You have to try A LOT of different models, always try to start with the simplest model and make it more complex as the data requires. For example: Start with a eins model and use only one eins temp. See where the model works and were it fails. In the regions that it fails add another eins temp...and so on. Shelly Kelly

2 1

Spline clamps
by SCalvin100＠aol.com 13 Jan '04

13 Jan '04

Hi all, Happy New Year to all! I've been busy pushing the new versions of Artemis and Ifeffit pretty hard (3-edge fits, over 100 paths, also over 100 guess,def,set parameters) and they seem to be behaving pretty nicely. There are certainly some advantages and some disadvantages to this compared to the old FEFFIT package; one major advantage is this will certainly reduced the learning curve for new undergraduates in my group. I am posting now because I am not sure I understand the behavior of the spline clamps in Ifeffit. When I set a top end to the spline range that is below the top of the data range, the high spline clamp has very little, if any, effect. (As a simple example, if I take the "clamp.prj" project that comes with Athena and set all of the spline ranges to top out at 400 eV, all four backgrounds look the same.) Is this what should be happening, or is this a bug? In case this is a bug, here's the version information: Athena 0.8.024 and ifeffit 1.2.4 running on Windows XP. I'll also be glad to send a project file to Bruce or Matt if they wish, although it would just be the clamp.prj file with the top of the spline range changed... --Scott

3 2

Feff Compatability
by Vivek S. Murthi 12 Jan '04

12 Jan '04

Hi Matt/bruce I would like to know if it is possible to run the Feff program on a Unix system. Right now I am running it on a PC which takes a long time to run. If it is possible, is there anything else I would need for Feff to run other than the executable and maybe the DOS compiler. Thanks vivek Vivek S. Murthi Graduate Research Assistant 419 Egan Research Center Northeastern Univeristy Boston, MA 02115 Phone: 617-373-8949

2 1

fitting questions
by k-kupiecki＠northwestern.edu 07 Jan '04

07 Jan '04

Hi, Just wondering if anyone out there can help answer a few questions I have.... I've been using Ifeffit/Artemis to try and fit to some experimental spectra.... and in trying to determine the best fitting parameters (k-range, r-range etc.)... I've come across a few things I'm not quite clear on..... (1) I was wondering (this might be a silly question) whether it was considered inappropriate to fit over a range larger than your spectra? (i.e. k-range of fit set to 2-15 when in actuality the spectra data only goes to ~12.5?) (2) I don't quite understand the following the significance of the following statement: These are not yet computed quite right in all situations... Chi-square for this data set = 9.01156 R-factor for this data set = 0.01651 When previously in the results the Chi-square and R-factor were given as: Independent points = 11.272460938 Number of variables = 11.000000000 Chi-square = 4.505779601 Reduced Chi-square = 16.537341618 R-factor = 0.016508545 Measurement uncertainty (k) = 0.007252615 Measurement uncertainty (R) = 0.091026879 Number of data sets = 1.000000000 I'm confused by the different Chi-square values, and what values of Chi square and R-factor for the data set would Artemis be satisfied with? (3) Also is there anyway to have Artemis deternine the best k-range (or perhaps a way to have it step through different values of kmin and kmax) ? Just wishful thinking on my part probably... Please let me know if you can help me understand these issues a bit better. Thanks!!!!!! Kristine Witkowski Northwestern University

3 2

RE: Ifeffit Digest, Vol 11, Issue 1
by Vivek S. Murthi 07 Jan '04

07 Jan '04

Hi Matt Earlier, I have also faced the problem faced by Tamas about the feff6 not running even though feff.inp is in the same folderas the executable. I was able to figure it out in my case. The problem was that the feff.inp file that we get from Tkatoms had to be for the right version of the Feff. There are different versions for the output availabe. If you trying running a feff.inp for feff8 format, it wouldnt run in feff6. I am not sure if tamas is also facing the same issue. thanks vivek Vivek S. Murthi Graduate Research Assistant 419 Egan Center Northeastern University Boston, MA-02115 Phone: 617-373-8949 On Tuesday, January 06, 2004 1:00 PM, ifeffit-request(a)millenia.cars.aps.anl.gov [SMTP:ifeffit-request@millenia.cars.aps.anl.gov] wrote: > Send Ifeffit mailing list submissions to > ifeffit(a)millenia.cars.aps.anl.gov > > To subscribe or unsubscribe via the World Wide Web, visit > http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit > or, via email, send a message with subject or body 'help' to > ifeffit-request(a)millenia.cars.aps.anl.gov > > You can reach the person managing the list at > ifeffit-owner(a)millenia.cars.aps.anl.gov > > When replying, please edit your Subject line so it is more specific > than "Re: Contents of Ifeffit digest..." > > > Today's Topics: > > 1. Re: maximum argument length? (Matt Newville) > 2. Re: running feff - a problem (Matt Newville) > > > ---------------------------------------------------------------------- > > Message: 1 > Date: Mon, 5 Jan 2004 17:45:25 -0600 (CST) > From: Matt Newville <newville(a)cars.uchicago.edu> > Subject: Re: [Ifeffit] maximum argument length? > To: XAFS Analysis using Ifeffit <ifeffit(a)millenia.cars.aps.anl.gov> > Message-ID: > <Pine.LNX.4.44.0401051739360.18361-100000(a)millenia.cars.aps.anl.gov> > Content-Type: TEXT/PLAIN; charset=US-ASCII > > Hi Peter, > > Sorry for the delay... Christmas/New Year's holidays. I'll be back > for real later this week. > > On Tue, 23 Dec 2003, Peter Southon wrote: > > > Is there a limit on the total length of arguments that follow an Ifeffit > > command? If so, is there a way of expanding the limit? > > Yes, there is a limit. Actually, there are several limits that can > affect how long the command line can be. The length limit on the > strings in the code are 2048 characters, but the strings get passed > back and forth enough times that I wouldn't claim that a single > command with 2047 characters will always work. More importantly, > there are various limits on different parts of the command length, > which is what I think you're seeing. > > > I've constructed a rather complicated model which uses 37 paths (mostly > > multiple scattering). Furthermore, quite a few of these paths are sensitive > > to scattering angle, so I have calculated FEFF files for three versions > > each of most paths (each for a different scattering angle), as suggested in > > the course "EXAFS Analysis with FEFF and FEFFIT" and the old Feffit manual. > > Thus the final Ifeffit command calls on a lot of FEFF paths, each > > specifically numbered according to path and angle. > > Unfortunately, after adding a few extra paths I get error messages when > > running the feffit command: "feffit:unknown key" followed by a list of > > paths about halfway through my list, and also "feffit: no chi(k) data > > array?". I assume that this is because ifeffit will only swallow half the > > command, which is: > > > > feffit(1-4,500,510,520,600,610,620,700,710,720,800,810,820, > > 9-13,15-16,1700,1710,1720,1800,1810,1820,1900,1910,1920,2000,20 10,2020, > > 37, 3800,3810,3820,3900,3910,3920,4000,4010,4020,4100,4110,4120, > > 4200,4210,4220,4300,4310,4320,4400,4410,4420,4500,4510,4520, > > 4600,4610,4620,4700,4710,4720,4800,4810,4820,4900,4910,4920, > > 5000,5010,5020,5100,5110,5120,5200,5210,5220,5300,5310,5320, > > 5400,5410,5420,5500,5510,5520, > > chi = data15.chi, macro = plot_fit) > > > > Short of actually renumbering all my paths so that they run sequentially, > > is there any solution...? > > Eek. This is my mistake, and I'll have to send repaired code for > this to work, which could be as soon as the end of this week. You're > hitting an accidentally low limit on the length of the 'path list'. > The string '1-4,500,...,5520' needs to be converted to a list of > integers: the limit on this string is currently only 256 characters, > which is definitely a mistake and will be increased to 1024 shortly. > > Another possible approach for this problem is to use the > unadvertised, "slightly magic" behavior that an undefined path is > silently ignored in the sum-over-paths. That is, if you define > paths 500, 510, 520, but not any of the path numbers in between, > using '500-520' would be the same as '500,510,520'. Though this > will give the same sum of paths, it _currently_ won't help you, as > it will quickly use up too many of the paths in the number of > internally stored paths used in the code. This will be fixed and > documented for the next version too (which should be within a couple > weeks), and the path indices will be checked to be between 1 and > 9999, so that saying '1-9999' will mean "use all defined paths". > > In principle, that could make the sort of problem you're dealing > with easier, but it might be real work for you at this point too. > > Sorry for all the trouble, and hopefully we can get it to work soon. > > --Matt > > > ------------------------------ > > Message: 2 > Date: Mon, 5 Jan 2004 17:47:16 -0600 (CST) > From: Matt Newville <newville(a)cars.uchicago.edu> > Subject: Re: [Ifeffit] running feff - a problem > To: XAFS Analysis using Ifeffit <ifeffit(a)millenia.cars.aps.anl.gov> > Message-ID: > <Pine.LNX.4.44.0401051745310.18361-100000(a)millenia.cars.aps.anl.gov> > Content-Type: TEXT/PLAIN; charset=US-ASCII > > Hi Tamas, > > > I'm trying to run feff to do the feff paths calculation the same way I've > > done before (feff file from TkAtoms and the executable feff6l are in the > > same folder), and now it just doesn't want to work. > > Hitting feff6l causes the opening of the DOS window, but it closes > > immediately and doesn't do the calculation. Does anyone know what's the > > cause of this? > > (I have the updates version of Ifeffit and the updater, too.) > > With 1.2.4(+ the updates), it should be that clicking on the feff6l batch > file brings up a dialog to browse for feff.inp, and then runs feff6l on > that file, in that directory. I think this batch file could be copied > from C:\Program Files\Ifeffit (or whereever Ifeffit is installed) and > still run OK. But I believe a shortcut to the executable (feff6l.exe) in > another directory will not work, as this executable needs to find DLLs to > run properly. That's why there are the batch files. > > If the proper environmental variables are set and the Ifeffit directory > is in your path, the executable should run from a DOS command window. > You could add this to autoexec.bat (or look at feff6l.bat): > SET IFEFFIT_DIR=C:\Program Files\Ifeffit > SET PATH=%PATH%;%IFEFFIT_DIR% > SET PGPLOT_DIR=%IFEFFIT_DIR% > SET PGPLOT_DEV=/GW > > or set these environmental variables from the Control Panel (Generally > Control Panel -> Settings -> Advanced -> Environmental Variables, though > it does vary with versions of Windows). Actually, for Feff6, only the > path needs to be set to include the Ifeffit directory. > > Please let me know if these suggestions don't help enough to get > fef6l working for you... > > --Matt > > > > ------------------------------ > > _______________________________________________ > Ifeffit mailing list > Ifeffit(a)millenia.cars.aps.anl.gov > http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit > > > End of Ifeffit Digest, Vol 11, Issue 1 > **************************************

2 1

RE: [Ifeffit] fitting questions
by Kelly, Shelly D. 07 Jan '04

07 Jan '04

Hi Kristine, I feel like I can add something to question number 3. Please see below > (3) Also is there anyway to have Artemis determine the best > k-range (or perhaps a way to have it step through different > values of kmin and kmax) ? Just wishful thinking on my part > probably... No, Artemis does not do this automatically. Here is a recipe that I use to determine the k-range. 1. back-up to the Athena stage. Look closely at the edge of the xmu(e) data and the background. Count the number of oscillations until you get to the flat region of the sepctra (about 30 eV away from the edge or so). Then plot the chi(k) data and count the oscillations to figure out where this node is in the chi data and set that to kmin. Usually the value for kmin is a node around 2 to 3 1/Angstroms. 2. Make about 6 copies of the chi(k) data. In each of the copies make sure that all the parameters are the same. Set kmin to the value from step 1. Then choose a node for kmax that looks really clean around 8 1/Angstroms. Then for the following copies choose the next and next nodes for kmax. Then compare the Fourier transform of all the copies. As you include more data the peaks should become more well defined but smooth. If you see a bunch of oscillations with a high frequency all over then you have gone too far for kmax. 3. Model the data with these conservative choices for the data range. Once you have a model that you like, try including another oscillation or two on both sides and see if the theory agrees. At this point you can also include a theory in the bkg subtraction step to get down to lower k. Shelly

1 0