Hi Steven,
On Tue, 27 Apr 2004, Steven Sangyun Lim wrote:
My question is how low can I set the k range? Is there any minimum range for reliable result? Please let me get out of this dilemma...
Scott and Matt both answered this question differently than I would. I guess I am very practical. If I was asking this question I would want a set of rules that would help me figure it out the k-range. So here one way to do that. 1) Look at the data in Athena to determine a good k-range to use. Here is how. 1a) As Matt suggested find a node in the chi(k) data that is around 3 inverse angstroms and set that to the kmin value. 1b) make about 6 copies of the chi(k) data set, all with the same kmin value. All the other FT values should be as you used in Artemis. (k-weight, window, dk) 1c) highlight each data set choosing a larger and larger node for kmax. Start really safe value around 8 inverse angstroms. Then pick the next node for the next copy and so forth using up all the data. For example, the first copy of the data set has a k-range [3.2 to 8.2] the next copy has k-range [3.2 to 9.3] and so forth....until the last has a k-range of [3.2 to 14] 1d) Plot the FT of the first three data sets all together (using the purple buttons to select them and the purple R button). Look at the data with the magnitude and the real or imaginary part. With a small k-range the FT signal will be very smooth curve in the magnitude, real and imaginary parts. You will tend to see two types of changes as you increase the k-range. 1) peaks will become more resolved, larger amplitude and thinner. --this is a good kind of change 2) eventually you will see noise in the FT of the data. The noise will look like high frequency pulses in the real and imaginary part of the spectra all over in R. From one data set to the next random signals will appear and disappear as you include noise in the data. --this is the bad kind of change. Keep comparing three data sets, drop the first one then add another with slightly larger k-range. You should be able to find the cross over from the good kind of a change to the bad kind of a change. 1e) By using this comparison you should be able to choose a k-range that is large but not too large. It should result in a conservative k-range to use as a starting point for your modeling. Be careful about the k-weight that you use when you make this comparison. It needs to be the same that you are using in Artemis to model the data. Larger k-weights will have smaller "good" k-ranges for the data. 2) Pull your data into Artemis. Use the good k-range that you determined from step one. 2b) Determine a model that works and look at the results including the uncertainties. 2c) Increase the k-range and compare the results. Try to decrease kmin and increase kmax. Do one thing at a time and see how the results vary. You will need to make a lot of comparisons, see how your results and uncertainties change as you adjust these values. 2d) Decide on a k-range that gives you the "best" results. 3) You may want to go back into Athena and use a standard to adjust the low k region of the chi(k) data, depending on how your FT looks at low R. This can help you stretch the kmin value. HTH Shelly
I have an addition to Shelly's "practical" advice: 2e) Try the fit at different k-weights. (If you were fitting at multiple k-weight simultaneously, try one k-weight at a time.) In my experience, with a reasonable k-range you may find the statistical quality of your fit (r-factor, chi-square) changes markedly, but the fitted parameters should not drift outside of the ranges defined by their uncertainties. If they do, it is either a sign that you should reduce the k-range or that you have a tenuous and unstable fit that should be viewed with considerable skepticism. --Scott Calvin Sarah Lawrence College At 04:03 PM 4/27/2004 -0500, you wrote:
2) Pull your data into Artemis. Use the good k-range that you determined from step one. 2b) Determine a model that works and look at the results including the uncertainties. 2c) Increase the k-range and compare the results. Try to decrease kmin and increase kmax. Do one thing at a time and see how the results vary. You will need to make a lot of comparisons, see how your results and uncertainties change as you adjust these values. 2d) Decide on a k-range that gives you the "best" results.
Shelly
Hi Shelly, Steven, Scott, Out of curiosity, how does kmax_suggest from chi_noise() or feffit() agree with the somewhat elaborate determination of a good value for kmax for noisy data that Shelly and Scott described? The hope is that kmax_suggest could give an idea for how far out the data is worth analyzing, but it hasn't been run through the mill too much. I don't know if artemis shows this value, so you might have to look in the Ifeffit buffer, maybe even typing show kmax_suggest after doing a fit. Thanks, --Matt
On Wednesday 28 April 2004 01:08 pm, Matt Newville wrote:
I don't know if artemis shows this value, so you might have to look in the Ifeffit buffer, maybe even typing show kmax_suggest after doing a fit.
That's the best way. "Show scalars" from the Show menu after a fit would also do the trick. I would be pleased to better use kmax_suggest in A&A. I am open to suggestions. B -- Bruce Ravel ----------------------------------- ravel@phys.washington.edu Code 6134, Building 3, Room 405 Naval Research Laboratory phone: (1) 202 767 2268 Washington DC 20375, USA fax: (1) 202 767 4642 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/
participants (4)
-
Bruce Ravel
-
Kelly, Shelly D.
-
Matt Newville
-
Scott Calvin