[Ifeffit] data after rebinning look worse in k-space than before?

Hi When it comes to rebinning and/or averaging, my preferred way is to use smoothing spline interpolation as implemented by IgorPro from Wavemetrics. With a little care, smooth curves with no artifacts can be obtained, in particular with oversampled data. We use it a lot with XPS data, where in my opinion the usual box averaging is really bad, producing lots of irrelevant wiggles in the base line, whereas smoothing splines do not have this problem. The reference for the algorithm is: "Smoothing by Spline Functions", Christian H. Reinsch, Numerische Mathematik 10. There are probably other implementations to do this. Michael.

Am 28.06.2018 um 20:20 schrieb Carlo Segre :

Hi Edmund and Chris:

I totally agree but for the same reason as Edmund, it is much more efficient to let the monochromator and not stop and start all the time. Rebinning is useful if you are sure that it works properly.

Carlo

On Thu, 28 Jun 2018, Edmund Welter wrote:

...

Dear Chris,

...

Rebinning is always fraught; I would normally not allow a student to do it... I thought you would say that! ;-) and I share your concern, actually that is the reason why we store ~8600 data points or more for a scan of 1200 eV. Don't mess with the raw data! But 8600 (or 22000...) is a significant oversampling. We could simply measure for a certain longer time per point, than read the counters and store the values. The results turned out to be less good, some weird technical reason. We could also sum up n points and divide the result by n, so that we would end with 8600/n data points in the final data file. In most cases that would certainly be ok, but information would be irretrievably lost.

But, at some point one has to reduce the number of points and make use of all the counts that were collected within a certain reasonable interval like 0.5 eV or 0.05 Ang-1. But it looks as if artifacts might arise from doing it twice, first explicitly by using Athena's rebin function and than again implicitly by mapping the data onto the 0.05 Ang-1 grid.

Cheers, Edmund