GArSoft Tracking Update Tom Junk DUNE HPGTPC Meeting January 15, - - PowerPoint PPT Presentation

GArSoft Tracking Update

Tom Junk DUNE HPGTPC Meeting January 15, 2019

Removing Outer Hits

Jan 15, 2019 Tom Junk | GArSoft Update 2

Field cage extends beyond the readout planes. Charge drifting at larger radii than readout is simulated to land on the

• uter row of pads.

(is this right?) Ends up distorting track ends ("feet"). Solution – just cut them

• ut.

Channel Mapping Upgrades

• Found bugs in the association of (y,z) positions with pads in

GArSoft

• Overflows and underflows in pad-row calculations on the sides
• f chambers are now protected
• Some "ghost" tracks now have vanished
• Hits that were displaced from their true locations now fixed –

usually occur where tracks cross boundaries between IROC and OROC, or between sectors.

• Tiny improvements in momentum resolution
• Raw Digit event displays now look better.

Jan 15, 2019 Tom Junk | GArSoft Update 3

Example Displaced Hit

Jan 15, 2019 Tom Junk | GArSoft Update 4

Hit-Finding Upgrades

Investigated why tracks at high ! were getting split. Traced this to deficiencies in the hit finder.

• Old: Waveform blocks over threshold are just called "hits".

Waveform storage is in blocks over threshold already, so the hit finder just calculated averages and RMS's.

• Problem with old hit-finder: If several tracks are close (near primary

vertex), or if the track points along E, the hits are very long and substructure is lost. Even if there is no substructure, giving the vector hit finder a small number of hits to work with produces vector hits with just one hit in them (no direction)

• New: Start new hits on a waveform if:
• They get too long (1 cm, fcl parameter)
• The waveform drops to 50% of the peak and then goes back up by at least

a factor of 1.3

Jan 15, 2019 Tom Junk | GArSoft Update 5

Definition of !

• Slope variable: s = sqrt(dy2+dz2)/dx diverges for isochronous

tracks

• Replace functionality with lambda = ACot(s).
• lambda = 0 for tracks perpendicular to the fields, ±"/2 for tracks

parallel to the fields.

Jan 15, 2019 Tom Junk | GArSoft Update 6

! E, B, x y z

! in pictures

• A pair of vector hits (blue) that should match up. They follow a

helix (dashed black). The average direction is shown in red

• A pair of vector hits that don't match up. The average direction

is shown in red.

Jan 15, 2019 Tom Junk | GArSoft Update 7

Pattern Recognition Upgrades

• Goal – reduce the fraction of tracks that are stitched incorrectly across

the primary vertex

• Split the ! test in two:
• Make the ! test two-dimensional, in the (y,z) plane (perpendicular to the B

field). Cartoon works for circles, not so much for sine waves.

• Now require " to match for two vector hits to be associated to each other.

Expect " not to vary along a helix.

• One source of ambiguity – vector hits have a two-fold ambiguity in which

way they point. " test needs to allow for this. Could define a sign for dx and constrain the sign.

• Also could add an !-like change in distance vs angle and separation but

requires some work and thought. Easy just to compare angles, and the cut

• n #" can be much tighter than the cut on 3D angle matching

Jan 15, 2019 Tom Junk | GArSoft Update 8

Muons in numuCC events: Efficiencies

Jan 15, 2019 Tom Junk | GArSoft Update 9

Momentum, Angle, and Position Resolution of Muons

Jan 15, 2019 Tom Junk | GArSoft Update 10

Was: ±5% !p/p resolution: Now 4.3%

Vertex performance

Jan 15, 2019 Tom Junk | GArSoft Update 11

Many more vertices found. (many more short tracks found near primary vertex) Many extra ones near busy primary vertices

All particle position, momentum, and angle resolutions in numuCC events

Jan 15, 2019 Tom Junk | GArSoft Update 12

Investigating Muon Momentum Resolution

• Would like to separate out effects of pattern-recognition failures,

fitting, and detector performance.

• Figured out how to minimize pattern-recognition failures: run

single muons! Much lower chance to associate wrong hits.

• !xz = 10∘ !yz = 75∘
• - almost perpendicular to B and E. Ran

1000 events each. Efficiency is ~100% (I think one was lost).

• All tracks start in the center of the detector (+ 10 cm in x)

Jan 15, 2019 Tom Junk | GArSoft Update 13

P (GeV/c) #P/P % 0.25 3.2 ± 0.3 0.5 3.1 ± 0.3 1.0 3.2 ± 0.3 2.0 3.1 ± 0.3 4.0 4.3 ± 0.4 8.0 6.3 ± 0.7

Gluckstern Formula

• I asked Justo if he could evaluate the momentum resolution

using the Gluckstern formula in the NDTF report, DUNE DocDB 1792:

Jan 15, 2019 Tom Junk | GArSoft Update 14

For your numerical check, I get the following: sigma = 1.E–3 m B = 0.4 T N = 200 L = 2 m X_0 = 2.6 m p = 1 GeV dp/p (mea) = 0.004 dp/p (msc) = 0.022

From Justo: I get 2.9% from msc using X_0 = 13 m. Independent

• f PT

To Do

• Trim the number of vertices found. More activity at primary -> many (fake) vertices found.
• Figure out how to keep the pattern recognition from stitching tracks across the primary vertex.

Partially addressed, but still a problem.

• Look at split tracks – partially addressed, but still split tracks in there
• Use the backtracker to make track completeness and purity metrics. Expect completeness to be

more of an issue, perhaps purity for very short stubs at the vertex.

• Address curlers.
• Address low-momentum tracks near the vertex: Got more of these, needs more work.
• Vertex finding measurement uncertainties
• Event display updates – vector hits, pickability, truth labels, etc.
• Extrapolate to ECAL
• Try with more complex events (pileup with particles coming from ECAL)
• Look at performance of hit clustering algorithm – already found the lowest-hanging fruit, but the

solution was quick and dirty.

• Get the CRY and Radiological generators working
• Write technote

Jan 15, 2019 Tom Junk | GArSoft Update 15

To Do: Address Photon Conversions and other Momentarily Coincident Tracks

• If a vector hit is associated with two others, then attach the two

groups together.

Jan 15, 2019 Tom Junk | GArSoft Update 16

A problem: this algorithm stitches together the legs of photon conversions and other momentarily coincident tracks

Improved identification of short tracks

Jan 15, 2019 Tom Junk | GArSoft Update 17

Too Many Vertices Near the Primary

Jan 15, 2019 Tom Junk | GArSoft Update 18

Leo's working on this. Issue of deciding which tracks to associate to vertices. Too many combinations are treated as distinct. Also a stray short track following a longer one adds confusion.

Momentarily Coincident Tracks

Jan 15, 2019 Tom Junk | GArSoft Update 19

Extras

Jan 15, 2019 Tom Junk | GArSoft Update 20