Malfonction of --use-bf-as-xsec for gundamCalcXsec when trying to reconstruct post-fit distribution.

[BenjouilleDem]

When using gundamCalXsec to reproduce the post-fit MC from gundamFitter there are still a few discrepancies between the distribution from gundamCalcXsec and from gundamFitter.

To compute them with gundamCalcXsec, I used in the config file UseBfAsCentralValue:True and nothing in the command. Indeed, when adding --use-bf-as-xsec in the command, the distribution doesn't match with post-fit MC from gundamFitter. One can see this problem in these plots :

Martini1pi_Full_PostFit_FHC FGD1 #nu_{#mu} CC 0#pi 0p 0#gamma.pdf
Martini1pi_Diff_Full_PostFit_FHC FGD1 #nu_{#mu} CC 0#pi 0p 0#gamma.pdf

The different points represent different configurations/commands used for gundamCalcXsec.
In the first plot :

• Cross : When using --use-bf-as-xsec in the command.
• Diamond : When using --use-bf-as-xsec in the command plus UseBfAsCentralValue:True in the config file.
• Triangle : When only using UseBfAsCentralValue:True in the config file.
• The red line is the post-fit MC from the Fitter.

In the second plot, the absolute differences between the different points and the red line are shown.

• Green line : When using --use-bf-as-xsec in the command.
• Yellow dashed line : When using --use-bf-as-xsec in the command plus UseBfAsCentralValue:True in the config file.
• Pink line : When only using UseBfAsCentralValue:True in the config file.

These tests have been made on 1.8.1 and the main branch of GUNDAM (on 19/07/2024) and gave similar results. (These plots are made with Martini1pi FDS and for the sample CC0pi0p0gamma)

So it seems that --use-bf-as-xsec doesn't work properly when trying to rebuild the post-fit distribution. Similarly, the “best result” (when UseBfAsCentralValue:True is used in the configuration file) still shows some discrepancies with the post-fit MC distribution.

[ClarkMcGrew]

My understanding is that this is fixed in the main branch. I did a quick check and it doesn't look like porting the "fix" back to 1.8.x is practical (unfortunately, there are a lot of class/file renaming changes between main and 1.8.x, so merging isn't possible, and it looks like this might need significant changes in 1.8.x).

At a recent working session meeting, we discussed the possibility of keeping the 1.8.x series (long term support) for the non-crosssection studies, and start freezing a second, slightly less stable, series for the cross section fitting.

We have testing to keep their results more or less in sync, but they won't be bug-for-bug compatible. We're still working on the detailed validation of 1.8.x relative to published results, but when that is finished, it should be fairly straight forward to validate main against 1.8.x (and port any fixes into main).

This comment about GUNDAM versions also applies to #531