This is a rather broad issue encompassing:
i). Merging existing Particles together to increase (most likely) shower completeness.
ii). Merging remnants with existing Particles to increase completeness.
iii). Bringing remnants into existence as their own Particles.
Speculative, longer-term: experiment with moving the shower branch growing into a 3x2D operation, with branch additions considered from all view at same time (currently additions are made completely independently).
#2 Updated by John Marshall about 3 years ago
- % Done changed from 30 to 80
Key issues affecting completeness of reconstructed showers:
1. Elements of sparse shower excluded - Appear as separate 3D Particles, or Remnant 2D Clusters (not in Particles)
2. Event Slicing (to tackle left-over cosmic rays) splits shower into multiple slices.
Use sliding linear fits to 3D Shower Clusters to define 3D cones:
-Use cones to aid collection of downstream fragments.
-Can use neutrino interaction vertex to define/choose cone direction.
-Need to be rather “generous” to collection most/all fragments.
-3D linear fit provides an axis; divide into layers. Assign Hits to layers and calculate layer “fit contributions”.
-By sliding/summing along nearby n layers, can obtain a local linear fit position and direction.
-Now exploit sliding 3D fit results to define cones:
--Fit position in chosen layer can provide cone apex
--Mean fit direction for next m layers defines direction
--Hit positions/algorithm chooses angle and length.
Exploit reusable 3DSlidingConeFit functionality in a number of algorithms and tools in reconstruction:
1. 3D Event Slicing: use cones to collect 3D Hits for more complete slices.
2. 3D Particle → 3D Particles: shower cone fits pick up downstream 3D fragment Particles.
3. 3D Particle → 2D Clusters: project cone fits, pick up downstream 2D remnant Clusters.
Also add simple backstop alg to assign any left-over “isolated” Clusters to nearest Particle within range.
Associated pattern recognition performance shown at MicroBooNE collaboration meeting.