Mattione Update 09042013
From GlueXWiki
Contents
CDC Track Finding - Basics
- Note: Some steps (and many details) are omitted entirely in this brief overview.
Build Super Layer Seeds
- In each ring, link adjacent hits together (DCDCRingSeed).
- In each super layer, link each possible combination of adjacent DCDCRingSeed's together to form super layer seeds (DCDCSuperLayerSeed).
- NEW: Do for both axial super layers and stereo super layers (previously just axial).
- NEW: Identify which DCDCSuperLayerSeed's contain potential spiral turns (where a spiraling track turns back inward/outward).
Build Track Circles
- HEAVILY MODIFIED: Link adjacent DCDCSuperLayerSeed's together to form tracks (DCDCTrackCircle).
- Previously, the axial DCDCSuperLayerSeed's were linked by progressively fitting track circles and selecting the next axial super layers that were close to the track fit.
- Start in super layer 1 (axial) and expand outwards towards super layer 7 (axial) (was previously in reverse direction)
- For each unique combination of axial super layers, create a new DCDCTrackCircle object.
- Keep track of which stereo DCDCSuperLayerSeed were used when linking the DCDCTrackCircle objects. There may be (many) more than one for each super layer.
Fit Track Circles & Truncate: Axial
- Fit the hits in the axial super layers (plus the beamline) to a circle with DHelicalFit.
- Reject tracks for which half of the axial hits are not close to the circle fit.
- NEW: Truncate and re-fit track circles that have picked up erroneous super layer seeds in the outer super layers.
- This is used to save (for example) forward-going tracks that pick up bad hits from another track (e.g. a large-angle track near in φ)
Filter Track Circles: Axial
- If two track circles share >= 50% of their hits, reject the DCDCTrackCircle with the worse circle-fit weighted-chisq/ndf (weighted by #axial super layers squared)
- NEW: If two track circles are spiral-linked in their last axial super layer, have opposite charges, and have similar circle centers, reject the DCDCTrackCircle with the worse circle-fit weighted-chisq/ndf.
Select Stereo Super Layer Seeds
- Previously, all stereo hits that intersected the circle and were nearby were lumped all together. Now, stereo super layer seeds are used instead.
- Project all hits from all possible stereo super layers onto the circle.
- For all possible combinations of stereo super layers:
- Do a linear regression fit of the projected stereo hits to calculate θ & Vertex-Z
- NEW: Select the combination of stereo super layer seeds that have the best chisq/ndf from the regression fit.
Filter Track Circles: Stereo
- NEW: Find super layer seeds that are shared between DCDCTrackCircle's, and remove them from all DCDCTrackCircle's except the one with the best weighted chisq/ndf from the regression fit.
- Like the axial truncation, this is used to save (for example) forward-going tracks that pick up bad hits from another track (e.g. a large-angle track near in φ).
- However, do NOT remove the stereo super layer if it's the last one on the track.
- NEW: If two track circles are spiral-linked in their last stereo super layer, have opposite charges, and have similar circle centers, reject the DCDCTrackCircle with the worse circle-fit weighted-chisq/ndf.
Refinement
- Find single, unused hits in the next (outward) super layer and add them to track circles if nearby.
- This primarily helps tracks that exit the CDC early (e.g. forward going).
- Redo circle fits, using both the new hits added just before, as well as the intersection points between the two stereo super layers (e.g. SL2 & SL3).
- Re-project stereo hits onto the new circle fits.
- Do a linear regression fit on a subset of the projected stereo hits to calculate θ & Vertex-Z
- SEMI-NEW: A subset is chosen to optimize the final fit results. This is because there are regions/cases for which the projected stereo-hit-positions are bad.
Finalize
- Calculate the position and momentum of the track at a point outside of the target and the start counter.
- This is because the track finding does not take energy-loss into account.
- Note that this skews the comparison between the DTrackCandidate and DMCThrown kinematics, since they are reported at different positions (r ~= 9.5cm, r = 0cm, respectively).
- Create DTrackCandidate objects.
Single π+ Track
Single π+ Reconstruction: Current (Trunk) Code, trackeff_hists plugin
- The below plots show the track reconstruction efficiency and resolution for the trunk using the single-track scripts plugin (trackeff_hists).
- However, this efficiency is extremely misleading:
- If a track is reconstructed using any of the hits that were on the thrown track it is considered "found."
- This is regardless of whether the reconstructed track momentum is anywhere close to the thrown values.
Single π+ Reconstruction: Current (Trunk) Code, NEW trackeffv2 plugin
- The below plots show the track candidate and time-based track reconstruction efficiency for the trunk using a new plugin (trackeff_hists).
- This plugin compares matches all of the thrown and reconstructed particles in the event, and requires that their momentum be "close."
- "Close" for track candidates is defined as: Δp/p < 20%, Δθ < 40 degrees, Δφ < 40 degrees, ΔVertex-Z < 1 km
- For thrown tracks with θ < 5 degrees, there is no cut on the candidate Δφ
- "Close" for time-based tracks is defined as: Δp/p < 10%, Δθ < 15 degrees, Δφ < 15 degrees, ΔVertex-Z < 10 cm
- "Close" for track candidates is defined as: Δp/p < 20%, Δθ < 40 degrees, Δφ < 40 degrees, ΔVertex-Z < 1 km
- Note that the degradation of the efficiency at forward angles matches the track resolution plot in the previous section.
- If I expand the requirements for the "Close" window, the efficiency increases, and the chosen parameters were somewhat arbitrary.
Single π+ Reconstruction: NEW Spiral Code
- The below plots show the track candidate and time-based track reconstruction efficiencies for the new spiral code.
- Note: The trackeffv2 plugin compares matches all of the thrown and reconstructed particles in the event, and requires that their momentum be "close."
- "Close" for track candidates is defined as: Δp/p < 20%, Δθ < 40 degrees, Δφ < 40 degrees, ΔVertex-Z < 1 km
- For thrown tracks with θ < 5 degrees, there is no cut on the candidate Δφ
- "Close" for time-based tracks is defined as: Δp/p < 10%, Δθ < 15 degrees, Δφ < 15 degrees, ΔVertex-Z < 10 cm
- "Close" for track candidates is defined as: Δp/p < 20%, Δθ < 40 degrees, Δφ < 40 degrees, ΔVertex-Z < 1 km
- Selected Track Candidate 1-D Kinematics Comparisons
- The below plots show the improvements in the track candidate and time-based track reconstruction efficiencies with the new spiral code.
γp→π+π+π-(n)
- The below plots show the track candidate and time-based track reconstruction efficiencies for both track finding codes, and the improvement.
- The definition of closeness is the same as before (see the single-track section).
- Events were generated at beam energy of 9 GeV and a t-slope of 5.0
Summary
- Overall track reconstruction improvement percentages for all tracks in this topology with momentum LESS than 3 GeV/c.
- WARNING: This is not the overall gain for the topology, as many pions have larger than 3 GeV/c of momentum (their overall gain will be smaller).
Particle | Track Candidates | Time-Based Tracks |
---|---|---|
π+ | 10.3% | 5.38% |
π- | 8.08% | 4.47% |
π+
- Thrown, Track Candidate 1-D Kinematics
- Efficiencies
Stage | Efficiency - Current Code | Efficiency - Spiral Code | Improvement | Improvement - Zoomed |
---|---|---|---|---|
Track Candidates | ||||
Time-Based Tracks |
γp→π+π+π-π-π0(p) (b1pi)
- The below plots show the track candidate and time-based track reconstruction efficiencies for both track finding codes, and the improvement.
- The definition of closeness is the same as before (see the single-track section).
- Events were generated with a beam energy of 9 GeV and a t-slope of 5.0
Summary
- Overall track reconstruction improvement percentages for all tracks in this topology with momentum LESS than 3 GeV/c.
- WARNING: This is not the overall gain for the topology, as many pions have larger than 3 GeV/c of momentum (their overall gain will be smaller).
Particle | Track Candidates | Time-Based Tracks |
---|---|---|
π+ | 7.93% | 5.21% |
π- | 12.5% | 6.23% |
p | 23.7% | 22.6% |
π-
- Thrown
- Efficiencies
Stage | Efficiency - Current Code | Efficiency - Spiral Code | Improvement | Improvement - Zoomed |
---|---|---|---|---|
Track Candidates | ||||
Time-Based Tracks |
Proton
- Thrown, Track Candidate 1-D Kinematics
- Efficiencies
Stage | Efficiency - Current Code | Efficiency - Spiral Code | Improvement |
---|---|---|---|
Track Candidates | |||
Time-Based Tracks |
γp→K+K+Ξ-
- The below plots show the track candidate and time-based track reconstruction efficiencies for both track finding codes, and the improvement.
- The definition of closeness is the same as before (see the single-track section).
- Events were generated with beam energies between 8.4 and 9 GeV
Summary
- Overall track reconstruction improvement percentages for all tracks in this topology with momentum LESS than 3 GeV/c.
- WARNING: This is not the overall gain for the topology, as some kaons have larger than 3 GeV/c of momentum (their overall gain will be slightly smaller).
Particle | Track Candidates | Time-Based Tracks |
---|---|---|
K+ | 21.7% | 14.1% |
π- | 35.1% | 28.7% |
p | 12.8% | 16.7% |
Sample Event
Reconstructed Hits | Truth + Reconstructed Hits | Current (09/03/13) Code Track Candidates | Spiral Code Track Candidates | |
---|---|---|---|---|
π-
- Thrown
- Efficiencies
Stage | Efficiency - Current Code | Efficiency - Spiral Code | Improvement | Improvement - Zoomed |
---|---|---|---|---|
Track Candidates | ||||
Time-Based Tracks |
K+
- Thrown
- Efficiencies
Stage | Efficiency - Current Code | Efficiency - Spiral Code | Improvement | Improvement - Zoomed |
---|---|---|---|---|
Track Candidates | ||||
Time-Based Tracks |
Proton
- Thrown
- Efficiencies
Stage | Efficiency - Current Code | Efficiency - Spiral Code | Improvement |
---|---|---|---|
Track Candidates | |||
Time-Based Tracks |
bggen (pythia)
- The below plots show the track candidate and time-based track reconstruction efficiencies for both track finding codes, and the improvement.
- The definition of closeness is the same as before (see the single-track section).
- Events were generated with beam energies between 8.4 and 9 GeV
Summary
- Overall track reconstruction improvement percentages for all tracks in this topology with momentum LESS than 3 GeV/c.
- WARNING: This is not the overall gain for the topology, as many pions have larger than 3 GeV/c of momentum (their overall gain will be smaller).
Particle | Track Candidates | Time-Based Tracks |
---|---|---|
π+ | 16.7% | 11.3% |
π- | 21.8% | 12.7% |
p | 12.4% | 6.92% |
π-
- Thrown
- Efficiencies
Stage | Efficiency - Current Code | Efficiency - Spiral Code | Improvement | Improvement - Zoomed |
---|---|---|---|---|
Track Candidates | ||||
Time-Based Tracks |
Proton
- Thrown
- Note this is in log-scale.
- Efficiencies
Stage | Efficiency - Current Code | Efficiency - Spiral Code | Improvement | Improvement - Zoomed |
---|---|---|---|---|
Track Candidates | ||||
Time-Based Tracks |
Hit Selector - bggen (pythia) Events
- New Code Only: Time-Based Track Efficiency improvements with hit selector disabled at all stages.
Protons | π+'s | π-'s |
---|---|---|