Open AsaNehm opened 2 months ago
Having a closer look at the iterations for calculating the track length reveals this
All entries after this event have track length 0 in all iterations. In reconstruction and truth!!!
(full output file at /exp/dune/app/users/pnehm/output_UV_NERSC.txt)
Checking for the truth for each iteration first if both hits are inside TMS (TMS_Geom::IsInsideTMS) seems to help
One thing I found is that the geometry gets corrupted somehow if it tries to measure a position far from the TMS. When that happens, all future calls are broken. This seems to be what's happening here. I fixed it by limiting the z in the world
The reason this happens with the nersc files specifically is because they have rock muons. The rock muons are more likely to reach this boundary where the corruption occurs
Still a smaller 0 bin for the truth remains...
For the full first NERSC samples 610 truth events have a density-weighted track length of exactly 0.
There are definitely also a non-negligible number of events where the reconstruction gets a considerable density-weighted track length while the truth stays at 0.
This seems to me like there is something bad going on with the truth
Track length [mm] vs density-weighted track length [g/cm^2] for reco (cut off at 6500mm)
and truth (cut off at 6500mm)
The performance plots for the track momentum/length reconstruction show a weird behavior![Reco_tracklength_full](https://github.com/DUNE/dune-tms/assets/101250148/46c162b6-ae68-40e3-8929-332f33b5efbe)