[Ifeffit] What differences are there between Athena and AUTOBK backgrounds?
Hi all, Today I tried making Athena perform exactly the same background subtraction as AUTOBK 2.61 (all parameters the same, working on the same data, using the same theoretical standard, etc.). I then let FEFFIT take the chi(k) data for each and perform a Fourier transform. When I did this, AUTOBK would always generate a very low r (< 0.2 angstroms) peak of significant amplitude that was nearly absent in the Athena background. There were other subtle differences as well, but that one stood out. Of course, a peak that low in r should have no effect on the fit, but it made me curious--were substantive changes in the background subtraction algorithm made when AUTOBK was incorporated into IFEFFIT? --Scott Calvin Sarah Lawrence College
Scott, There are some differences between autobk 2.61 and Ifeffit's spline() command. I consider autobk 2.61 to be pretty old (it's from 1995): I found a version 2.63 (1996) this evening: I can probably find older versions, but I don't have complete records of all changes to these codes (autobk 2.0 and feffit 2.0 were from 1993, and there were 1.0 versions!). I do fondly remember autobk 2.0 and feffit2.05 (the first to really do multiple-data-sets!) but I'm pretty sure I wouldn't want to run them. Anyway, enough nostalgia: the main differences between Ifeffit and autobk earlier than 2.92 (year 2000, at which point autobk became part of ifeffit) are: 1. fixed an 'off by k=0.05Ang (1 grid point)' error. 2. changes in E->k interpolation schemes: now using the 'rolling boxcar' average discussd here earlier this year. 3. switch to fully double precision. Minor changes were made in other pieces: methods for choosing the spline knots, initial guesses for the free parameters, the actual spline evaluation, Fourier transforms, and fitting. There are some differences in normalization, too. I think none of these are significant changes, but the 'noise level' is pretty high for background subtraction, making it hard to tell if a change make things better or worse. You might look if there is an 'off by one grid point' difference between the chi(k) spectra. --Matt
participants (2)
-
Matt Newville -
Scott Calvin