Hit merging: second bumps in Q and E distributions

[YifanC]

Currently the hit merger (https://github.com/DUNE/ndlar_flow/blob/develop/src/proto_nd_flow/reco/charge/calib_hit_merger.py) yields a second bump in Q and E distribution towards higher energy deposition in the calib_final_hits. This shoulder feature appears in both the data files and the simulation files and seems to be unphysical. (https://indico.fnal.gov/event/63652/contributions/286004/attachments/175752/238516/20240227_2x2_M1_reflow_checkup.pdf.

I think the current hit merger (5febf1c) only uses the time differences of packets on the same pixel. Perhaps we can judge based on charge whether they should be merged or not. Extending to the truth backtracking/labeling, perhaps in the simulation, we could label "induction" and "collection" charge. If necessary, we can investigate the merge within neighbouring pixels as well.

[krwood]

I'm not sure exactly what is meant by "unphysical" here, but the distributions don't really surprise me. Given we are merging "prompt" hits to create "final" hits here, it makes sense that we get additional peaks at multiples of the one in "prompt" hits.

That being said, the hit merging algorithm definitely has much room for improvement. Even without improving or changing the algorithm, the configuration parameters for the existing algorithm can be optimized - this hasn't been done before. For that, I would suggest starting with the simulation and using the truth backtracking to characterize the performance (with some figure of merit like hit location residuals) of the hit merging, and then scanning different configuration parameters to optimize some FOM.

[YifanC]

Maybe "unphysical" is stronger than it intended to be. I would say it has unknown impact on the segment dE/dx and it indicates something (I can not describe it better than that). If or how it's angular dependent, charge dependent etc. If you were summing up all charge/energy for a given cluster, I think it doesn't matter. However, it's one way to ease the downstream reconstruction, but also an anchor for us to study the cause of these "hits". Does Pandora for example not care about the calorimetric substructure of a cluster? (I don't know the answer.)
Although the above may not be constructive on how to carry out is work, this is my opinion why we are doing it.