Debug different background results cmp to old analysis

[Vindaar]

Currently we observe a quite different background rate spectrum compared with Christoph's old analysis. While it does produce similar results above \SI{2}{\kilo \electronvolt}, below we "lose" a lot of background (would be great of course). Interesting things to take in mind / to try out:

• [ ] loss of background (\le \SI{2}{\kilo \electronvolt}) for gold region, but increase for whole chip!
• [ ] difference of calibration factors for carge calibrated Fe spectrum of old vs new framework
• [ ] gas gains are different, use the mean of the histogram instead of the fit result as the gas gain. Just multiply bin center * bin content. Check whether to use raw histogram or histogram based on fit result?
• [x] check all cuts applied to reference datasets. Maybe application of one is wrong? Maybe value of one is wrong? Especially for the small energy bins?
• [x] plot the reference datasets / the resulting distributions after the cuts and compare with old plots (see his PhD thesis)
• [ ] calculate properties from single raw event and see if and which properties show different values than event from old framework. Might be done using the reference tree, since there we should have everything in one place?
• [ ] plot again the distributions from the actual data after reconstruction and compare
• [x] take a look at PhD thesis, reference spectrum bin E has a lot of plots to compare against
• [ ] or calculation of logL spectra of reference sets check for overflow bins, empty bins in the resulting spectra etc., check the values used to construct these distributions (number of bins, start and end value etc). If several bins were empty in the low energy ranges and we'd get an event from the background dataset whose logL value resides in that bin, it'd drop out! That would fit with our observation!
• [x] check result if exact cut values from Christoph's PhD thesis is used. -> no difference after change in 1st comment.
  • [x] check if cut values are slightly different due to binning -> seems to be the case. Christoph's values don't exist in my binning. Typically my code chooses the lower bin compared to Christoph's value, with the exception of the C-0.6kV dataset, in which the next bin would fit exactly. Might have to do something about this in another issue.
• [ ] what else?

[Vindaar]

A major bug was fixed in https://github.com/Vindaar/seqmath/commit/fa243404b2da2ddf974bfb074ef1120538151b56, which caused us to get the wrong cut values for the logL values, since we dropped the empty beginning bins for the logL reference distributions. https://github.com/Vindaar/TimepixAnalysis/commit/a7e20c9ebd41a14a355111ad6af29ab20c115679 and https://github.com/Vindaar/TimepixAnalysis/commit/7a5401d251fa5a47db705c8acd4803fd91e4b105 also helped debugging so far and to an extent by changing the number of bins for the logL histograms.

[Vindaar]

Note for myself. After major bug in previous commit:

• introduced cuts used on X-ray spectra being created (see C. Krieger PhD appendix B & C) and create spectra from silver plus gold region
• add cleaning cut on transverse RMS: <= 1.5

[Vindaar]

Closed as a "won't fix". Lots of work went into this, but this is irrelevant for this repository.