Hi Han Sen: On Sat, 18 Sep 2010, Han Sen Soo wrote:
My first objective is to get an idea of the coordination number of Ti, and I'm trying to do this in Artemis by using the parameter N as my S0^2 and letting the amplitude of S0^2 vary. I'm not sure if this is appropriate?
I think that if you attached the project file it would be easier to understand this question and whether you are doing the "right thing".
A problem I am encountering is that the physically reasonable fits look visually unacceptable. I am just trying to fit the first Ti-O shell initially and plan to work on the 2nd shell subsequently. The R space fit for the first shell has 2 peaks that are about 0.3 Å apart at 1.2 and 1.5 Å (uncorrected). With only 1 shell of O atoms at a distance I found for similar materials, a reasonable fit was found (including background fitting) but the amplitude of the fit at the 1.5 Å peak is too high while the amplitude at the 1.2 Å peak is too low. I naively tried to improve the fit by adding a 2nd shell at a shorter distance and tried different combinations of distances 0.02 Å apart, but I consistently find that the sigma^2 values for one or both shells become negative. I also realized that the resolution of the data may not be sufficient to distinguish shells that are less than 0.2 Å apart, which is what I am expecting based on data on similar materials. I have attached a pdf image of one of the bad fits.
I think that the biggest problem oyu have is that you are fitting the background to remove the big peak at 0.85A. As you can see from your curve, the background extends up into the actual data and is undoubtedly skewing your results. Not only that but the background fitting introduces 6 additional parameters and thus artificially is making the fit "better". I would advise you to stay away from the background fitting and instead try to improve the background subtraction to remove the low-R peak. can you tell us what were your settings for the background subtraction? If you can't get rid of the background peak with any other means (there are other tips on the IFEFFIT Wiki), you can always just fit the data and ignore the low-R peak. That would be preferable than distorting the fit. Cheers, Carlo -- Carlo U. Segre -- Professor of Physics Associate Dean for Graduate Admissions, Graduate College Illinois Institute of Technology Voice: 312.567.3498 Fax: 312.567.3494 segre@iit.edu http://www.iit.edu/~segre segre@debian.org