[Ifeffit] MoS2 fit

Grant Bunker bunker at agni.phys.iit.edu
Thu Nov 11 12:12:58 CST 2004


I haven't played with these data or this analysis, but the difficulty in
fitting the known structure may suggest a problem with the data.

One experimental problem that springs to mind, since MoS2 is a layered
structure, is whether there might be a preferred orientation in the sample
particles, i.e. the particles might not be randomly oriented. Since
the x-ray beam is polarized this could cause some serious weirdness in the
XAFS fitting (if the polarization weren't explicitly considered).

grant

On Thu, 11 Nov 2004 ifeffit-request at millenia.cars.aps.anl.gov wrote:

> Send Ifeffit mailing list submissions to
> 	ifeffit at 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 at millenia.cars.aps.anl.gov
>
> You can reach the person managing the list at
> 	ifeffit-owner at 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. Commercial MoS2 fit (dmc at pdx.edu)
>    2. Commercial MoS2 -- Correction  (dmc at pdx.edu)
>    3. Re: Commercial MoS2 -- Correction  (Scott Calvin)
>
>
> ----------------------------------------------------------------------
>
> Message: 1
> Date: Thu, 11 Nov 2004 08:12:34 -0800
> From: dmc at pdx.edu
> Subject: [Ifeffit] Commercial MoS2 fit
> To: XAFS Analysis using Ifeffit <ifeffit at millenia.cars.aps.anl.gov>
> Message-ID: <1100189554.41938f7244a49 at webmail.pdx.edu>
> Content-Type: text/plain; charset=ISO-8859-1
>
> Hi List,
>
>      I have tried a variety of different things to get this fit to work.  It
> seems pretty decent now, although, for my two MS paths (4,5) I had to force the
> sO^2's to be positive.  In the process one of them is now zero.  My
> parameter 'amp' wants to be 0.5 so I had to set it to 0.7 because I know that
> it shouldn't be as low as 1/2.  Changing the k-weight doesn't seem to help.
> Here's my Data:
>
>                    Thanks, Dan
>
>
>
> Project title :  Fitting merge 10_15_04.chi
> ============================================================
>
> Independent points          =      28.581054688
> Number of variables         =      13.000000000
> Chi-square                  =    5995.284754934
> Reduced Chi-square          =     384.780419245
> R-factor                    =       0.043668737
> Measurement uncertainty (k) =       0.000187262
> Measurement uncertainty (R) =       0.006509968
> Number of data sets         =       1.000000000
>
>
> Guess parameters +/- uncertainties  (initial guess):
>   e0  =    0.8645560   +/-      1.2283680    (guessed as 0.864556 (1.228344))
>   ss  =    0.0041070   +/-      0.0005190    (guessed as 0.004107 (0.000519))
>   ss2 =    0.0043590   +/-      0.0005200    (guessed as 0.004359 (0.000520))
>   ss3 =    0.0217810   +/-      0.0333390    (guessed as 0.021781 (0.033341))
>   ss4 =    0.0000000   +/-      0.0509500    (guessed as 0.000000 (0.050951))
>   ss5 =    0.0000360   +/-      0.0063810    (guessed as 0.000036 (0.006382))
>   ss6 =    0.0028730   +/-      0.0071340    (guessed as 0.002873 (0.007133))
>   alpha =   -0.0012530   +/-      0.0031940    (guessed as -0.001253 (0.003193))
>   alpha2 =   0.0075160   +/-      0.0022250    (guessed as 0.007516 (0.002225))
>   alpha3 =  -0.0171580   +/-      0.0323200    (guessed as -0.017158 (0.032321))
>   alpha4 =  -0.0165410   +/-      0.1054840    (guessed as -0.016541 (0.105487))
>   alpha5 =  -0.0181860   +/-      0.0187940    (guessed as -0.018186 (0.018795))
>   alpha6 =   0.0227330   +/-      0.0147690    (guessed as 0.022733 (0.014767))
>
> Def parameters:
>   delr_1          =    -0.0057180
>   delr_2          =     0.0342980
>   delr_3          =    -0.0782970
>   delr_4          =    -0.0754820
>   delr_5          =    -0.0829880
>   delr_6          =     0.1037370
>
> Set parameters:
>   amp1            =  .7
>
>
> Correlations between variables:
>          ss3 and alpha4     -->  0.8462
>           e0 and alpha      -->  0.8335
>          ss4 and alpha3     --> -0.6693
>           e0 and alpha2     -->  0.6159
>           ss and ss2        -->  0.6114
>       alpha5 and alpha6     --> -0.5959
>       alpha4 and alpha5     --> -0.5821
>          ss4 and ss5        --> -0.5505
>          ss3 and alpha5     --> -0.5246
>        alpha and alpha2     -->  0.5146
>          ss5 and ss6        --> -0.4931
>          ss2 and alpha5     -->  0.4681
>          ss5 and alpha3     -->  0.3885
>          ss6 and alpha6     -->  0.3700
>       alpha2 and alpha5     -->  0.3208
>          ss2 and alpha4     --> -0.2992
>          ss4 and alpha5     -->  0.2955
>          ss3 and ss4        -->  0.2923
>          ss5 and alpha2     --> -0.2842
>           e0 and alpha4     --> -0.2781
>          ss2 and ss3        --> -0.2677
>       alpha3 and alpha5     --> -0.2523
> All other correlations are below 0.25
>
>
>
> ------------------------------
>
> Message: 2
> Date: Thu, 11 Nov 2004 08:20:02 -0800
> From: dmc at pdx.edu
> Subject: [Ifeffit] Commercial MoS2 -- Correction
> To: XAFS Analysis using Ifeffit <ifeffit at millenia.cars.aps.anl.gov>
> Message-ID: <1100190002.419391328eb1a at webmail.pdx.edu>
> Content-Type: text/plain; charset=ISO-8859-1
>
> Hi List,
>
>        It appears that changing the k-weight on my fit does improve the fit.
> I still have the problem of forcing the sO^2's on the two MS paths to be
> positive.  Here's some new data with k-weight 3:
>
>              Thanks, Dan
>
> Project title :  Fitting merge 10_15_04.chi
> ============================================================
>
> Independent points          =      28.581054688
> Number of variables         =      13.000000000
> Chi-square                  =    1527.118835332
> Reduced Chi-square          =      98.011262136
> R-factor                    =       0.025169471
> Measurement uncertainty (k) =       0.000185138
> Measurement uncertainty (R) =       0.081594539
> Number of data sets         =       1.000000000
>
>
> Guess parameters +/- uncertainties  (initial guess):
>   e0 =     1.1335810   +/-      1.3393990    (guessed as 1.133581 (1.339426))
>   ss =     0.0035870   +/-      0.0003110    (guessed as 0.003587 (0.000311))
>   ss2 =     0.0042850   +/-      0.0002780    (guessed as 0.004285 (0.000278))
>   ss3 =     0.0303100   +/-      0.0510420    (guessed as 0.030310 (0.051043))
>   ss4 =     0.0000000   +/-      0.0303840    (guessed as 0.000000 (0.030385))
>   ss5 =    -0.0008090   +/-      0.0043530    (guessed as -0.000809 (0.004353))
>   ss6 =     0.0059560   +/-      0.0089810    (guessed as 0.005956 (0.008981))
>   alpha=    -0.0006040   +/-      0.0023740    (guessed as -0.000604 (0.002374))
>   alpha2=     0.0073900   +/-      0.0016520    (guessed as 0.007390 (0.001652))
>   alpha3=   -0.0167250   +/-      0.0535940    (guessed as -0.016725 (0.053595))
>   alpha4=   -0.0120340   +/-      0.0773040    (guessed as -0.012034 (0.077302))
>   alpha5=   -0.0216900   +/-      0.0127530    (guessed as -0.021690 (0.012751))
>   alpha6=     0.0219980   +/-      0.0183950    (guessed as 0.021998 (0.018395))
>
> Def parameters:
>   delr_1          =    -0.0027560
>   delr_2          =     0.0337230
>   delr_3          =    -0.0763210
>   delr_4          =    -0.0549150
>   delr_5          =    -0.0989780
>   delr_6          =     0.1003830
>
> Set parameters:
>   amp1            =  .7
>
>
> Correlations between variables:
>           e0 and alpha      -->  0.8490
>           e0 and alpha2     -->  0.8006
>          ss3 and alpha4     -->  0.7123
>        alpha and alpha2     -->  0.6792
>          ss4 and alpha5     -->  0.6298
>           ss and ss2        -->  0.6114
>       alpha5 and alpha6     --> -0.5959
>          ss5 and alpha4     -->  0.5328
>          ss3 and ss4        -->  0.5300
>          ss5 and ss6        --> -0.4931
>          ss4 and alpha3     --> -0.4617
>          ss2 and alpha5     -->  0.3847
>          ss6 and alpha6     -->  0.3700
>          ss2 and alpha3     -->  0.3276
>          ss4 and ss5        --> -0.2764
>       alpha2 and alpha4     -->  0.2506
> All other correlations are below 0.25
>
>
>
>
>
> ------------------------------
>
> Message: 3
> Date: Thu, 11 Nov 2004 12:19:06 -0500
> From: Scott Calvin <scalvin at slc.edu>
> Subject: Re: [Ifeffit] Commercial MoS2 -- Correction
> To: XAFS Analysis using Ifeffit <ifeffit at millenia.cars.aps.anl.gov>
> Message-ID: <3.0.1.32.20041111121906.00a657e0 at mail.slc.edu>
> Content-Type: text/plain; charset="us-ascii"
>
> Hi Dan,
>
> A couple of quick thoughts, that might also be helpful to people starting
> out with EXAFS fitting.
>
> First of all, the good news is that changing the k-weight does not seem to
> affect the values ifeffit finds for the parameters significantly (i.e. I
> didn't see any that moved outside of their uncertainties). That's very
> reassuring. The fact that the fit quality is noticeably better at k-wt 3
> suggests to me that you might try raising kmin a bit--maybe the k-wt 3 fit
> works better because the fits are struggling at the low end of the k-range?
> Sometimes it's hard to get the background to work right down at low k, and
> it looks like you have plenty of data range based on the number of
> independent points.
>
> Second, I don't like forcing S02 to 0.7 because 0.5 is too low very much.
> An iffy sample or problems with normalization can both suppress S02 below
> the nominally ideal value. Forcing the fit to adopt a higher S02 then just
> has the effect of screwing up correlated variables, like the Debye-Waller
> factors. I'd look for explanations for the low S02 rather than just
> disallowing it. (I certainly have set S02 in some of my published work, but
> those were cases of complex fits where it seemed to be a false minimum,
> rather than a sample or normalization error.)
>
> Finally, I'd try some "reasonable" constraints on your MS paths. If they
> are triangle paths, it is not possible to come up with a rigorous formula
> for how to write the debye-waller factors and alpha's for MS paths in terms
> of the direct scattering paths, because you don't know for sure how the
> motions of nearby atoms correlate. But in my experience, which is mainly
> with metals and transition metal oxides, fits are generally not
> particularly sensitive to the details of MS path constraint strategies.
> Here's a constraint strategy of medium complexity that I teach my
> undergrads that you can try:
>
> Def the alpha's of the MS paths to be the average of the alpha's for the
> direct scattering paths that make them up, weighted by the length of the
> segment. It's quite probable in a material such as yours that some segment
> will be a S-S path which you don't have available as a direct scattering
> path. Maybe just leave that piece out of the average.
>
> Def the ss's of the MS paths to be the sum of the ss's for the direct
> scattering paths that make them up, divided by 4. Why divided by 4? Because
> the ss for a half-path is rigorously equal to 1/4 the ss for a whole path.
> Why add them? If the ss of each segment is truly uncorrelated, then adding
> them is statistically proper. Of course, this is a truly awful
> approximation in some cases, notably as the paths come closer to being
> focused (i.e. one of the angles of the triangle becomes large). In cases
> where the biggest angle in the triangle is more than about 140 degrees, I
> just treat it as a focused path, but I doubt that's true for your two MS's.
>
> Please understand that this constraint scheme is fairly arbitrary! When
> working with a new material, I suggest trying a constraint scheme like
> this, and then seeing how sensitive the fit is to this constraint. For
> example, multiply all the ss's of the MS's paths by 2 and see how stable
> the fit is. Go back to allowing the alpha for the MS paths to vary freely
> and see again if the fit is stable. This kind of probing will tell you
> whether the MS paths are something that have to be carefully considered, or
> if they are just a small correction to your fit that should be included but
> don't have to be included particularly carefully. It is my experience that
> the latter is often the case, but you have to find out for yourself on this
> material.
>
> --Scott Calvin
> Sarah Lawrence College
>
> At 08:20 AM 11/11/2004 -0800, you wrote:
> >Hi List,
> >
> >       It appears that changing the k-weight on my fit does improve the fit.
> >I still have the problem of forcing the sO^2's on the two MS paths to be
> >positive.  Here's some new data with k-weight 3:
> >>
>
>
> ------------------------------
>
> _______________________________________________
> Ifeffit mailing list
> Ifeffit at millenia.cars.aps.anl.gov
> http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit
>
>
> End of Ifeffit Digest, Vol 21, Issue 8
> **************************************
>




More information about the Ifeffit mailing list