20/03/2025 14:30
Performance of pattern finding (reminder):
Sean and I met to discuss the problems with the pattern finder. Our current idea is that there are two main failing points:
Tracklets line fitting being skewed causing improper distance calculation
Example:
Truth:
Reconstruction:
Potential Solution:
In general, we need a better fitter. How to go about that (and at what stage) is unclear. One option is to only do linear fits near the stop and start of tracks (preventing "pull" from scattering). Another option is to have a more robust fitter that considers scattering like seen above.
Particles (like Bhabas) being created far from tracklet endpoints
Example
Truth:
Reconstruction
Potential Solution:
First, it's unclear whether or not these electrons will be accurately resolved by the tracklet finder (or if they even need to be).
Assuming these electron tracklets are accurately resolved, we instead need to project tracklet starts onto some finite space (ex. line segment). That way, the distance calculation will be based on how close the new tracklets start is to an existing tracklet as a whole, not just the endpoints of another tracklet. If this is more computationally expensive, it can be done as a "second pass", i.e. the original method of just comparing endpoint distances can be done first.
20/03/2025 14:41
Some discussion points for Jessie:
- How much info from tracklet finding (all hits vs some fit)
- If we have a fit, how much information are we getting from that?
- (maybe for Patrick) how much computatation resources can we spare? I.e. how fast does this need to be.