Hey Dan, I only meant if you use the same alpha for all paths, you're assuming an isotropic expansion. Different alphas is one way to get rid of that. One other thing I noticed-- in your first message, it says the r-range you used in the fit is the default-- 1

Hi Shelly,

I seperated the delr's with different alpha terms, but Mike seems to think that this is assuming isotropic expansion that isn't there. Also, the reduced chi-square has gone up 150 or so.

delr# define: alpha# * reff

Thanks for the help, Dan

Results with 1 eO and differing delr's:

Independent points = 14.312500000 Number of variables = 7.000000000 Chi-square = 6186.771083042 Reduced Chi-square = 846.054165202 R-factor = 0.031090393 Measurement uncertainty (k) = 0.000181935 Measurement uncertainty (R) = 0.004860095 Number of data sets = 1.000000000

Guess parameters +/- uncertainties (initial guess): ss = 0.0042720 +/- 0.0006590 (guessed as 0.004636 (0.000411)) e0 = 1.3293700 +/- 1.4673640 (guessed as 1.765958 (1.154388)) alpha = -0.0001750 +/- 0.0040010 (guessed as 0.001090 (0.002843)) ss2 = 0.0043180 +/- 0.0007370 (0.0000) ss3 = 0.0072300 +/- 0.0164390 (0.0000) alpha2 = 0.0076940 +/- 0.0027880 (0.0000) alpha3 = -0.0512670 +/- 0.0232140 (0.0000)

Def parameters: delr_1 = -0.0006930 delr_2 = 0.0305320 delr_3 = -0.2034430

Set parameters: amp = 0.7

Correlations between variables: e0 and alpha --> 0.8441 ss3 and alpha2 --> 0.7651 ss2 and alpha2 --> -0.6441 e0 and alpha2 --> 0.5682 ss2 and ss3 --> 0.4871 alpha and alpha2 --> 0.4785 ss3 and alpha3 --> 0.3825 e0 and alpha3 --> -0.2743 All other correlations are below 0.25

Original (2 e0s):

Independent points = 14.312500000 Number of variables = 6.000000000 Chi-square = 5778.957783654 Reduced Chi-square = 695.212966455 R-factor = 0.029041008 Measurement uncertainty (k) = 0.000181935 Measurement uncertainty (R) = 0.004860095 Number of data sets = 1.000000000

Guess parameters +/- uncertainties (initial guess): e0_1 = -4.0347670 +/- 2.9998980 (guessed as -4.072009 (1.919404)) ss = 0.0042860 +/- 0.0005990 (guessed as 0.004636 (0.000411)) e0 = 1.6770760 +/- 1.3230290 (guessed as 1.765958 (1.154388)) alpha = 0.0009180 +/- 0.0034050 (guessed as 0.001090 (0.002843)) ss2 = 0.0048160 +/- 0.0006910 (0.0000) ss3 = 0.0225310 +/- 0.0657250 (0.0000)

Def parameters: delr_1 = 0.0036440

Set parameters: amp = 0.7

Correlations between variables: e0_1 and alpha --> 0.8542 e0 and alpha --> 0.8470 e0_1 and e0 --> 0.7407 e0_1 and ss3 --> 0.7060 alpha and ss3 --> 0.4715 ss2 and ss3 --> -0.4214 e0_1 and ss2 --> -0.3892 e0 and ss3 --> 0.3772 alpha and ss2 --> -0.2680 All other correlations are below 0.25

_______________________________________________ Ifeffit mailing list Ifeffit@millenia.cars.aps.anl.gov http://millenia.cars.aps.anl.gov/mailman/listinfo/ifeffit

Hi Dan, You can't exclude paths by pushing the R-range up to just short of the Reff for the path! This is for three reasons: 1) The EXAFS Fourier transform is not the radial distribution function, although it is "related" to the radial distribution function. One example of this is that the contribution from a path tends to be centered lower (maybe half an angstrom) in R-space than the Reff associated with the path. 2) Atoms vibrate, or have static disorder. That's what sigma2 is about, after all. This means that if the mean absorber-scatterer distance is x, there are many cases where it is somewhat less than x. 3) Technical effects having to do with taking a Fourier transform of a finite data range introduce additional broadening into the signal due to a given path, so that it extends well below and above its mean value. So how do you know how high to go in R? There are many ways to decide. One is to include the paths that you don't want to worry about in the fit (e.g. the MS paths at 3.989). Then, when the fit is done, use Artemis to plot those paths. You can then visually see how far down in R they have a noticable effect, and set your Rmax accordingly. If you really want to be sure. run a fit with them included and one without. If the fit does not change significantly (R factor, parameters stay pretty stable), then you know you're OK. If the fit does change significantly, you've got to lower Rmax. --Scott Calvin Sarah lawrence College

Dear List,

Thanks for all the good suggestions. Here's how my fit stands at this point. I have a few multiple scattering paths at 3.989, in order to excude these 2 I put the R-range from 1-3.975. It seems that you think I should include these paths, if so, do I treat them a little differently. They don't seem to help the fit using the same parameter strategy. Additionally, I extended the k-range from 2-13.5 to 2-15. This seemed to improve the reduced chi-square dramatically.