Download lori_stevens_0507

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Identical particles wikipedia , lookup

Elementary particle wikipedia , lookup

ATLAS experiment wikipedia , lookup

Compact Muon Solenoid wikipedia , lookup

Transcript
Efforts to Improve the Reconstruction
of Non-Prompt Tracks with the SiD
Lori Stevens
UCSC
ILC Simulation Reconstruction
Meeting
May 15, 2007
Includes contributions from:
Tyler Rice
Outline
Tyler Rice’s efficiency and purity results
from study of Tim Nelson’s
AxialBarrelTrackFinder algorithm
Z Segmentation algorithm (ZSeg.java)
Results after implementing ZSeg.java
Tyler’s phi-restriction and results (including
ZSeg.java implementation)
Detector
pythiaZPolebbbar-0-1000_SLIC_ v1r9p3_sidaug05.slcio,
without effects of beamsstrahlung or brehmsstrahlung
AxialBarrelTrackFinder1.java
Track pattern recognition using only the 5 outer
tracking layers
Works from outside inward (VXDCheater.java
has already removed hits from “prompt” tracks
originating within 20mm of the origin)
ABTF1 begins by using all sets of isolated hits in
3 layers to find circles that pass within 10cm of
the interaction point (original version had 1cm)
When 3 seed track is found, remaining layers
are checked for nearby hits
Event and Particle Requirements
Event (“Jet Accept Test”):
1. Cosine of thrust angle < 0.5
2. Thrust value > 0.94
“Findable” Particles:
1. Final State or Intermediate State with r
origin < 400mm and path length > 500mm
2. Transverse momentum > 0.75GeV
3. Carries a charge
4. |Cosine theta| < 0.8
5. Not backscatter off of the calorimeter
Particle and Track Definitions
“Found” Findable MC Particle: associated track
has “purity” >= 0.75 (at least 3 of 4 hits from
same MCP or at least 4 of 5 from same MCP)
“Missed” Findable MC Particle: all other findable
MC Particles
Fake track:
1. no majority MC Particle associated with track
2. tracks with bad purity (too few hits from
same MCP)
Tyler’s Results (have already been
presented in Beijing)
118/1000 events passed Jet Accept Test
304 total MC Particles
Efficiency:
131 Found with 5 hits (43%)
100 Found with 4 hits (33%)
73 Missed (24%)
Fake rates:
327 Fake (326/327 are 4 hit tracks: this implies
that 4 hit tracks cannot be used)
Digression: Other Studies by Tyler
(why is reconstruction efficiency not 100%?)
First Tyler tried requiring that particles hit each of
the 5 outer detector layers once and only once
Fewer candidate particles:
166 Findable MC Particles (304 before
requirement)
Efficiency:
113 Found with 5 hits (68% vs. 43%)
25 Found with 4 hits (15% vs. 33%)
28 Missed (17% vs. 24%)
Tyler’s Three-Hit Seed Study
Then Tyler also required all hits from three-hit seed
tracks to be associated with the same MC
Particle
166 Findable MC Particles (304 before
requirement)
Efficiency:
144 Found with 5 hits (87% vs. 43%)
15 Found with 4 hits (9% vs. 33%)
7 Missed (4% vs. 24%)
Motivation for Z Segmentation
Improve efficiency for finding MC Particles
(can we clean up the 3 hit seeds?)
Decrease number of 4 hit fake tracks: see
if we can make 4 hit tracks useable
Z Segmentation Algorithm (1 of 2)
ZSeg.java creates segmentation of z-axis into separate
modules (length set by user)
Algorithm is capable of offsetting individual layers,
amount set by user (still testing)
ZSeg.java contains ZCheckerExt method that takes
three SimTrackerHit arguments; this method is called
from inside AxialBarrelTrackFinder1.java
Method calculates minimum and maximum coordinates
of the z module for each of the 3 hits
Straight lines in r-z are projected from modules in layers
containing 1st two hits onto layer containing 3rd hit
Z Segmentation Algorithm (2 of 2)
Algorithm checks if 3rd hit is in a module consistent with
the1st two hits
For now, testing consistency in 3 hit seeds only (later to
include check for 4th and 5th hits)
Eventually algorithm will take in a list of hits and check
all possible 3 hit combinations for consistency, including
a test for whether to use extrapolation or interpolation
(currently using only extrapolation)
Original (Tyler’s) result: only require that hits are on
same side in z. This is not required when using z
segmentation.
Module Projection (extrapolation)
Note: No actual spacing between modules
Hit 1
Hit 2
Possible modules for following hits
Module Projection (interpolation)
Note: No actual spacing between modules
Hit 1
Possible
modules
Hit 2
Z Segmentation Results
Tyler
30cm
segments
10cm
segments
5cm
segments
1cm
segments
# MCPs
304
302
302
302
302
Found
with 5 hits
131
122
138
145
148
Found
with 4 hits
100
100
104
109
99
Missed
73
80
60
48
55
Fake
327
455
332
267
127
Implementing ZSeg.java
(“preliminary” results)
Z Segmentation Results
Number of tracks
500
30cm
450
400
10cm
350
Found: 5 hits
5cm
300
Found: 4 hits
250
1cm
200
Missed
Fake
150
100
50
0
0
0.5
1
Log base 10 of segmentation length
(in cm)
1.5
Another Idea: Require Hits to be in
Same Sector in Phi
Recall Tyler saw that a lot of inefficiency
and fake tracks due to bad 3 hit seeds
Clean up seeds by requiring that the phi
coordinate of all hits must be within π/2 of
each other
Also apply criterion to all hits once track is
found
Tyler’s Phi Restriction Results
(only require hits on same side in z)
304 total MC Particles
Efficiency:
145 Found with 5 hits (48%)
112 Found with 4 hits (37%)
47 Missed (15%)
Fake rates:
158 Fake (all 4 hit)
Tyler’s Results: New vs. Old
New
results
% of
MCPs
Old
results
% of
MCPs
# of MCPs
304
100%
304
100%
Found with 5 hits
145
112
47
48%
37%
15%
131
100
73
43%
33%
24%
Found with 4 hits
Missed
Fake
158
327
Z Segmentation Results
(Phi-Restricted)
Tyler
30cm
segments
10cm
segments
5cm
segments
1cm
segments
# MCPs
304
302
302
302
302
Found
with 5 hits
145
142
147
152
152
Found
with 4 hits
112
113
114
110
101
Missed
47
47
41
40
49
Fake
158
202
142
108
45
ZSeg.java with Phi Restriction
Phi Restricted Z Segmentation Results
Number of tracks
250
30cm
200
1cm
5cm
10cm
Found: 5 hits
150
Found: 4 hits
Missed
100
Fake
50
0
0
0.5
1
Log base 10 of segmentation length (in
cm)
1.5
Comparing Z Segmentation with
and without Phi Restriction
30cm 30cm
phi rest.
10cm 10cm
phi rest.
5cm
phi rest.
5cm
1cm
phi rest.
1cm
#MCPs
302 302 302 302 302 302 302 302
Found
with 5 hits
142 122 147 138 152 145 152 148
Found
with 4 hits
113 100 114 104 110 109 101
99
47
Missed
Fake
80
41
60
40
48
49
55
202 455 142 332 108 267
45
127
Graphs of Z Segmentation with and
without Phi Restriction
Z Segmentation Results
450
450
of
Found: 4 hits
400
350
300
Num ber
Found: 5 hits
tr acks
500
of
500
250
200
150
30cm
5cm 10cm
1cm
Missed
Fake
100
50
0
Num ber
t rac k s
Phi Restricted Z Segmentation Results
30cm
400
10cm
350
Found: 5 hits
300
Found: 4 hits
250
1cm
200
5cm
Missed
Fake
150
100
50
0
0
0.5
1
Log base 10 of segmentation length (in
cm)
1.5
0
0.5
1
Log base 10 of segmentation length
(in cm)
1.5
Comments on Results
Might expect 30cm segmentation to be
worse than simply requiring all hits to be
on same side of the detector
Assumption that tracks are straight in r-z is
less valid for low pT
Shortcomings
Projection in r-z will actually curve;
ZSeg.java treats track r-z projection as if it
were a straight line
Using ZPole; qqbar at 500GeV would be
even more challenging. Would like to
study but will need 500GeV qqbar with no
beamsstrahlung or brehmsstrahlung
For the Future
Check for z consistency in 4th and 5th hits
Take in list of hits and check all possible
hit combinations (including interpolation/
extrapolation check)
Check validity of straight line
approximation as a function of pT
The End