The Tool-based Reconstruction Algorithm for Characterising ...€¦ · The sample used •...
Transcript of The Tool-based Reconstruction Algorithm for Characterising ...€¦ · The sample used •...
The Tool-based Reconstruction Algorithm
for Characterising Showers*Dom Brailsford
FD sim/reco Meeting16/09/19
*Primary authors Dom Barker
Ed Tyley
Shower characterisation• Shower characterisation == the bit you do
after the pattern recognition
• Shower reconstruction is hard, particularly in a LArTPC
• The difficulty is frustrating because you primarily only need to know a few key pieces of information for physics analyses:
• Start position
• Initial direction
• Energy
• dE/dx
• *ID of the initial track stub
!2
Not another shower characterisation module?
• The Tool-based Reconstruction Algorithm for Characterising Showers (TRACS) is a shower characterisation module
• Developed by Dom Barker and Ed Tyley as part of the SBN shower reconstruction WG
• TRACS takes a recob::PFParticle as input and outputs a recob::Shower
• Just like PandoraShower and EMShower
• The key difference is TRACS outsources all of the characteristic calculations to a set of art tools where a tool has a specific purpose
• A tool to calculate the shower’s direction
• A tool to find the initial track stub’s hits
• A tool to calculate the initial track’s dE/dx
!3
Quick technical details
• You specify an arbitrarily sized list of the tools in FCL to configure the shower characterisation
• The only caveat is enough tools need to be run to characterise start position, direction, energy and dE/dx*
• *All current dE/dx need an initial track stub to function
• Every tool is exposed to the PFParticle under consideration, the art::Event and the ShowerElementHolder
• ShowerEleHolder holds the characteristics that each tool calculates, it can also hold an arbitrary number of transient objects as well as art data products destined for the ARTROOT output file
!4
Tools to be shown today• Shower start position
• ShowerStartPosition_tool: Takes the PFParticle’s vertex position as the start position (a la pandoraShower)
• Initial track stub finder
• Shower3DTrackHitFinder_tool: Tags space points contained in a cylinder whose axis is the calculated shower direction and starting at the calculated shower start position
• Initial track fitter
• ShowerPMATrackFinder_tool: Uses the Projection Matching Algorithm to form a track using the found initial track hits
• Shower direction
• ShowerPCADirection_tool: Applies a charge-weighted PCA to the shower space points; the primary axis is the shower direction (similar to pandoraShower)
• ShowerTrackTrajectoryPointDirection_tool: Takes the starting direction of the fitted track stub (recob::Track::DirectionAtPoint(1))
• dE/dx
• ShowerStandardCalodEdx_tool: Calculates a per-view dE/dx using the initial track hits. The tool calculates a per-view pitch by projecting the calculated shower direction into each view
!5
The sample used• Vertex-like particle gun
• Electron (0 < p < 4.5 GeV)
• Pi+ (0.5 GeV < p < 2.0 GeV)
• Fixed start pos (180,-300,580)
• Isotropic angle
• Downstream fcls:
• standard_g4_nooptical_dune10kt_1x2x6.fcl
• Disables the light sim
• standard_detsim_dune10kt_1x2x6.fcl
• standard_reco_dune10kt_nu_1x2x6.fcl
• For assessing the recob::Shower parameters, I only consider recob::Showers with the highest number of hits in an event AND which are also truth matched to the primary electron
!6
Quick aside: Pandora neutrino vertex position displacement
• | vertex_reco - vertex_true|
• Nice sharp peak at 0 cm
• Very broad second peak at ~120 cm
• Displacement is within 10 cm 34.6% of the time
0 50 100 150 200 250 300 350 400 450 500Dist. between true and reco. vertex (cm)
0
2000
4000
6000
8000
10000
12000
14000
16000
No.
eve
nts
!7
Mis-vertexing examples
Reco vertex
True vertex
Reco vertex
True vertex
!8
Mis-vertexing examples
Reco vertex True
vertex
Reco vertex
True vertex
!9
Number of reco. showers
• The number of ALL recob::Showers in each particle gun event
• Total stats: 49700
• 16.4% of particle gun events do not contain a recob::Shower
0 1 2 3 4 5 6 7 8 9No. reco. showers
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
No.
eve
nts
!10
Completeness/hit purity
33024 (66.4%) events contain at least one reco. shower truth matched to the primary electron
15110 (30.4%) events contain a primary electron-matched shower where the initial track stub contains at least one primary electron hit
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 shower with max No. hits)-Completeness (e
0
200
400
600
800
1000
1200
1400
1600
No.
eve
nts
hit-Initial track stub contains at least one primary e
matched showers-All primary e
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 shower with max No. hits)-Hit purity (e
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
No.
eve
nts
hit-Initial track stub contains at least one primary e
matched showers-All primary e
!11
*Shown for showers which contain the max no. of hits AND truth match to the primary electron
Low completeness eventsWire/tick views Ortho3D
displayOnly
spacepoints are shown
Clusters are matched by
colour
Black spots are unused hits
!12
Low completeness events
100120140160180X300−250−200−150−100− 50−Y 450
500
550
600
650
700
Z• Chain of trip ups
• Pat. rec. separates track stub and cascade bulk (pink line [track] attached to black splodge [shower])
• Start position of cascade bulk then placed at wrong end (green square)
• The track stub finder fails to find any initial hits
• Shower is thrown away as it is incomplete
• One of the much smaller showers (blue splodges) becomes the shower with the highest hits
!13
TRACS event display (shows space points)
A shower
A different shower
A track
Start pos.
Direction vector
Low completeness eventsWire/tick views Ortho3D
displaySpacepoints
and true trajectories
shown
Clusters are matched by
colour
Black spots are unused hits
Initial electron missing
space points
!14
Low completeness events• Another chain of trip ups
• Pat. rec. merges charged pion track with the whole electron shower
• Start position then picked to be the very end of the shower
• Charge-weighted PCA is biased towards centre of the pion track
• Track stub finder misses the shower space points as direction is mis-aimed
• Shower is thrown away due to being incomplete
• One of the much smaller showers (blue splodges) becomes the shower with the highest hits
110120130140150160170X350−300−250−200−150−Y
440
460
480
500
520
540
560
580
600
620Z
!15
TRACS event display
Shower start position
0 10 20 30 40 50 60 70 80 90 100Distance between true and reco. shower start (cm)
210
310
No.
eve
nts
hit-Initial track stub contains at least one primary e
matched showers-All primary e
• Using ShowerStartPosition_tool
• Takes the PFP’s vertex as the shower start position
• Plot shows the distance between the true and reco shower start positions
• Nice sharp spike at 0 cm in all cases
• Wide tail is largely eliminated for showers in which the track stub contains a primary hit
!16
*Shown for showers which contain the max no. of hits AND truth match to the primary electron
The initial track stub finder
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1Fraction of primary-matched hits
10
210
310
410
No.
eve
nts• Shower3DTrackHitFinder_tool
• IDs track space points as those enclosed in a cylinder at the start of the shower
• Plot shows the fraction of shower-contained primary electron hits that are part of the initial track stub
• Not the same as completeness
• 45% of initial track stubs contain a non-zero fraction of primary electron hits
!17
*Shown for showers which contain the max no. of hits AND truth match to the primary electron
1− 0.8− 0.6− 0.4− 0.2− 0 0.2 0.4 0.6 0.8 1 (no units)trued • recod
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
No.
eve
nts
'DirectionAtPoint' initial track direction tool
PCA direction tool
1− 0.8− 0.6− 0.4− 0.2− 0 0.2 0.4 0.6 0.8 1 (no units)trued • recod
0
2000
4000
6000
8000
10000
12000
No.
eve
nts
'DirectionAtPoint' initial track direction tool
PCA direction tool
DirectionPrimary electron showers with
max. no. hitsInitial track contains one primary
electron hit
!18
• PCA outperforms the direction from the initial PMA track in all cases
• This study will be redone with a new tool which uses pandora’s sliding linear fit once it becomes available
*Shown for showers which contain the max no. of hits AND truth match to the primary electron
dE/dx
0 1 2 3 4 5 6 7 8 9 10dE/dx (MeV/cm)
0
200
400
600
800
1000
1200
1400
1600
No.
eve
nts
hit-Initial track stub contains at least one primary e
matched showers-All primary e• ShowerStandardCalodEdx_tool
• Shower direction (required to calculate pitch) calculated by PCA tool
• MIP peak sharpens for cases where the track stub contains a primary electron hit
!19
*Shown for showers which contain the max no. of hits AND truth match to the primary electron
Good recos
Summary and future• The Tool-based Reconstruction Algorithm for Characterising Showers (TRACS)
is a shower characterisation module
• Developed by Dom Barker and Ed Tyley as part of the SBN shower reconstruction WG
• Relies on pandora PFParticle input
• TRACS outsources all calculation to an arbitrary set of art tools, each of which calculates a specific shower parameter
• TRACS is now being configured for use in the DUNE FD
• Future tools are under development/will soon be available
• Fitting the initial track stub with pandora’s sliding linear fit
• An incremental fit-based track stub finder!21
Backups
1− 0.8− 0.6− 0.4− 0.2− 0 0.2 0.4 0.6 0.8 1 (no units)trued • recod
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
No.
eve
nts
'DirectionAtPoint' initial track direction tool
PCA direction tool
What causes the spike at 0