GPU-based Online Tracking Algorithms & Application to D±→K∓π±π±

Mitglied der Helmholtz-Gemeinschaft
1
RUB Seminar Talk
3 February 2014, Andreas Herten
GPU-based Online Tracking
Algorithms &
Application to D±→K∓π±π±

View Slide

Outline
• Online Tracking for PANDA
• GPUs
• Online Tracking Algorithms
– Triplet Finder
– Hough Transforms
• D±→K∓π±π±
– EvtGen
– Hitcounting
– Tracking Benchmark
– Event Reconstruction
2

View Slide

PANDA &
ONLINE TRACKING
3

View Slide

PANDA — The Experiment
4
13 m
p
p

View Slide

PANDA — The Experiment
4
13 m
p
p
Magnet
STT
MVD
GEMs
FTS

View Slide

PANDA — Event Reconstruction
• Continuous read out
– Novel feature
– Background & signal similar
– No hardware trigger based on few sub-detectors,
but online event reconstruction using full detector information
5
(Reject background events, save interesting events)
Reduction
Amount:
Time:
~1/1000
50 ns/evt
Storage space for
oﬄine analysis
3 PB/y
Event:
Raw data:
2 × 107/s
200 GB/s
Rate Full version

View Slide

PANDA — Read Out Scheme
6

View Slide

Requirements to
Online Tracking
• Fast
• Sophisticated
algorithms possible;
reprogrammable
• Parallelism beyond
single devices
• Speed ↔ precision
6

View Slide

Requirements to
Online Tracking
• Fast
• Sophisticated
algorithms possible;
reprogrammable
• Parallelism beyond
single devices
• Speed ↔ precision
6
GPUs

View Slide

GPUS
7

View Slide

8
CPU & GPU
• Originally used for computer games
• Programming available as
General Purpose GPU, GPGPU
StarCraft II

View Slide

8
CPU & GPU
CPU GPU
• Originally used for computer games
• Programming available as
General Purpose GPU, GPGPU
StarCraft II

View Slide

CPU & GPU
9
GPU
CPU

View Slide

CPU & GPU
9
GPU
CPU
a1
→ b1
→ c1
a2
→ b2
→ c2
a3
→ …
parallel
a1
→ b1
→ c1; a2
→ b2
→ c2; a3
→ …
serial

View Slide

Challenges of GPU programming
10
=
Leveraging GPU
processing power
Writing adequate
algorithms adequately
parallel • non-blocking • computing-intensive
Getting data to GPU
integration into DAQ • high-level data
+

View Slide

11
ALGORITHMS #1
Triplet Finder
Hough Transform

View Slide

12
Triplet Finder
• Algorithm specifically designed for the
PANDA Straw Tube Tracker (STT)
http://www.fz-juelich.de/ias/jsc/
1.5 m
• Ported to GPU
– CUDA, Dynamic Parallelism, Thrust
– Quality of tracks comparable to CPU
Drift tubes: t0 needed
for drift circles
→ r(tArrive-t0)
1cm
Resolution without t0: (0.1 cm)
Resolution with t0: (0.015 cm)

View Slide

13
Triplet Finder
• Idea: Use only subset of detector as seed
– Don‘t use STT isochrones (drift times)
• Features
– Fast & robust algorithm, no t0 needed
– Many tuning possibilities
More

View Slide

14
Triplet Finder — Times
K20X

View Slide

Triplet Finder — Optimizations
• Bunching Wrapper
– Hits from one event have similar timestamps
– Combine hits to sets (bunches) which occupy GPU best
15

View Slide

15
Hit

View Slide

15
Hit Event

View Slide

15
Hit Event
Bunch

View Slide

15
Hit Event
Bunch
(N2) → (N)

View Slide

16
Triplet Finder — Bunching
K20X
Performance

View Slide

17
Triplet Finder — Kepler vs. Maxwell
Performance for diﬀerent GPUs
K20X
Maxwell
Performance: 1.3 TFLOPSsingle
Price: 130 €
750 Ti
Kepler
Price: 3600 €

View Slide

ALGORITHMS #2
18
Triplet Finder
Hough Transform

View Slide

Algorithm: Hough Transform
• Basic Hough idea: Generate lots of independent track
parameters; extract best-matching one.
• Three Hough Transforms:
– Line Hough Transform
• Around point-like hits (MVD, STT-prefit)
• Around isochronuous hits (STT)
– Circle Hough Transform (MVD, STT, GEM, …)
19
x
y
x
y
Hough Transform — Princip
→ Bin
giv
r
α

View Slide

Hough Transform — Variations
20
• Circle Hough with isochrones (xC, yC)ij = (xi + sij cos φj, yi + sij sin φj)
2 sij = ρi
2 - xi
2 - yi
2 / (xi cos φj + yi sin φj + ρi)

View Slide

-300 -200 -100 0 100 200 300
-300
-200
-100
0
100
200
300
HoughHisto_1
Entries 171941
Mean x -3.39
Mean y 14.01
RMS x 101.6
RMS y 123.5
0
2
4
6
8
10
12
14
16
18
HoughHisto_1
Entries 171941
Mean x -3.39
Mean y 14.01
RMS x 101.6
RMS y 123.5
21

View Slide

-300 -200 -100 0 100 200 300
-300
-200
-100
0
100
200
300
HoughHisto_1
Entries 171941
Mean x -3.39
Mean y 14.01
RMS x 101.6
RMS y 123.5
0
2
4
6
8
10
12
14
16
18
HoughHisto_1
Entries 171941
Mean x -3.39
Mean y 14.01
RMS x 101.6
RMS y 123.5
21
π+
π-
π-
K-
K+
π+

View Slide

-300 -200 -100 0 100 200 300
-300
-200
-100
0
100
200
300
HoughHisto_1
Entries 171941
Mean x -3.39
Mean y 14.01
RMS x 101.6
RMS y 123.5
0
2
4
6
8
10
12
14
16
18
HoughHisto_1
Entries 171941
Mean x -3.39
Mean y 14.01
RMS x 101.6
RMS y 123.5
22

View Slide

Circle Hough — GPU Performance
23
i5 @ 3.3 GHz
Variation of number of threads (and blocks): GPU vs. CPU
GTX 580

View Slide

Circle Hough — GPU Performance
23
i5 @ 3.3 GHz
Variation of number of threads (and blocks): GPU vs. CPU
GTX 580
~20 ×

View Slide

Circle Hough — Challenges & Todo
• CPU:
– Benchmarking; cut parameters
– Code cleanup
• GPU:
– Performance w/o PandaRoot
– More parallelism (hit → set of hits)
– Data transfer
• Tracking: Data or computing intensive?
• Diﬀerent types of memories
• Asynchronous transfer & processing
24

View Slide

GPUs — Summary
• Triplet Finder: Performance-optimized
– 14 µs / event
• Circle Hough: PandaRoot-integrated
• GPUs: Feasible for part of PANDA‘s online event
reconstruction system
– Also important: Data transfer to GPU
25
PndCircleHoughTask * chTracker = new PndCircleHoughTask();
chTracker-‐>AddHitBranch("MVDHitsStrip");
chTracker-‐>AddHitBranch("STTHit");
chTracker-‐>SetUseGpu(true);

View Slide

D± → K∓ π± π±
26
Description & EvtGen
Detector Response
Tracking Performance
Event Reconstruction

View Slide

D± → K∓ π± π±
27
Detector Response

View Slide

4 Proceedings of the DPF-2009 Conference, Detroi
Mass (MeV)
3000
3500
4000
4500
Charmonium Spectrum

?
K
c
J/\
K'
c
\'
\(4040)
\(4415)
\(4160)
h
c
F
c0
F
c1
F
c2
F'
c2
\''
13S
1
23S
1
13D
1
11S
0
21S
0
31S
0
41S
0
11P
1
21P
1
13P
1
23P
1
23P
0
13P
0
13P
2
23P
2
13F
2
11D
2
13D
2
13D
3
23D
3
23D
2
21D
2
33S
1
23D
1
43S
1
33D
1
DD
DD*
53S
1
Y(4260)
Z(3940)
Y(4660)
Y(4350)
Y(4008)
Z+(4430)
Z
1
(4050)
X(3872)
X(3940)
Y(3940)
Z
2
(4250)
X(4160)
Y(4140)
D
s
D
s
D*D*
D
s
D*
s
D*
s
D*
s
DD
1
D
s
D
s1
, D*
s
D
s0
D*D
2
D*
s
D
s2
Y(4630)
Figure 5: The Charmonium spectrum. The solid lines
are quark mode predictions [21] the shaded lines are the
observed conventional charmonium states [22], the hori-
zontal dashed lines represent various D(∗)
s
¯
D(∗)
s
thresholds,
and the (red) dots are the newly discovered charmonium-
like states placed in the column with the most probable
spin assignment. The states in the last column do not fit
elsewhere and appear to be truly exotic.
Figure 6
in B →
mass di
the B+
backgro
total un
4143.0
3.7 Me
JP C =
argue t
in B →
re 1.
(Left) Invariant mass of the 3⇡ system for 0.1 GeV/c2 < t0 < 1.0 GeV2/c2 (histogram), and intensity of the background wave
a flat distribution in three-body phase space (red triangles), obtained from a partial wave analysis in 40 MeV/c2 bins of the 3⇡ mas
escaled to the binning of the histogram (from [1]). (Right) Intensities of major waves (a) 1++0+⇢⇡ S, (b) 2 +0+ f2⇡ S, (c) 2++1+⇢⇡,
as the intensity of the exotic wave (d) 1 +⇢⇡ P, as determined in the fit in mass bins (data points with error bars). The lines represen
sult of the mass-dependent fit (from [1]).
.
Ds
Ds
*
D
Ds(2317)
(2317)
Ds(2460)
(2460)
Ds(2710)
(2710)
Ds (2860)
(2860)
J
Ds(3040)
(3040)
Observed
Ds1
Ds2
Mass (MeV/c )
3000
3500
4000
4500
?
c
J/
'
c
'
(4040)
(4415)
(4160)
h
c
c0
c1
c2
'
c2
''
13S
1
23S
1
13D
1
11S
0
21S
0
31S
0
41S
0
11P
1
21P
1
13P
1
23P
1
23P
0
13P
0
13P
2
23P
2
13F
2
11D
2
13D
2
13D
3
23D
3
23D
2
21D
2
33S
1
23D
1
43S
1
33D
1
53S
1
Y(4260)
Z(3940)
Y(4660)
Y(4350)
Y(4008)
Z+(4430)
Z
1
(4050)
X(3872)
X(3940)
Y(3940)
Z
2
(4250)
X(4160)
Y(4140)
D D
D D
Y(4630)
2
DD
DD*
DD
DD
D D
D D
D D
D D
s s
*
*
s s
*
D D
D D
s s
*
*
D D
D D
1
D D , D D
D D , D D
s s1 s s0
*
D D
D D
*
2
D D
D D
*
s s2
s s
re 2.
(Left) The Ds
meson spectrum as predicted by Godfrey and Igsur [12] (solid line) and by Di Pierro and Eichten [13] (dotted li
rimental values are shown by points; black points refer to old measured states, red ones to newly discovered. (Right) The charmoni
rum. The solid lines are CQM predictions [12], the shaded lines are the observed conventional charmonium states, the horizontal
ed lines represent various thresholds. The red dots are the newly discovered charmonium-like states placed in the column of the mo
able spin assignment. The states in the last column do not fit elsewhere and appear to be truly exotic.
ng list of states, with mass in the range 1–2 GeV/c2,
which an exotic interpretation has been claimed.
been found, allowing us to check the hyperfine contributi
to the q ¯
q potential. Taking the PDG11 [4] value for
28
D Mesons Spectrum
Godfrey (2009) »Topics in Hadron Spectroscopy in 2009«
arXiv:0910.3409
Predictions
Observed
New Discoveries
Many D(S) and old/new charmonium states decay via D meson
→ Good reconstruction important!
Predictions
Observed
Gianotti (2012) »Results and perspectives in hadron spectroscopy «
Phys. Scr. 2012 014014

View Slide

29
D Hadronic Decay
K∗
2
(1430)0 → K− π+
K− π+ e+ νe nonresonant < 7 × 10−3 CL=90% 864
K− π+ µ+ νµ ( 3.8 ±0.4 ) % 851
K∗(892)0 µ+ νµ ,
K∗(892)0 → K− π+
( 3.52±0.10) % 717
K− π+ µ+ νµ nonresonant ( 2.0 ±0.5 ) × 10−3 851
K− π+ π0 µ+ νµ < 1.6 × 10−3 CL=90% 825
π0 e+ νe ( 4.05±0.18) × 10−3 930
ηe+ νe ( 1.14±0.10) × 10−3 855
ρ0 e+ νe ( 2.18+0.17
−0.25) × 10−3 774
ρ0 µ+ νµ ( 2.4 ±0.4 ) × 10−3 770
ωe+ νe ( 1.82±0.19) × 10−3 771
η′(958)e+ νe ( 2.2 ±0.5 ) × 10−4 689
φe+ νe < 9 × 10−5 CL=90% 657
Fractions of some of the following modes with resonances have already
appeared above as submodes of particular charged-particle modes.
K∗(892)0 e+ νe ( 5.52±0.15) % 722
K∗(892)0 µ+ νµ ( 5.28±0.15) % 717
K∗
0
(1430)0 µ+ νµ < 2.4 × 10−4 CL=90% 380
K∗(1680)0 µ+ νµ < 1.5 × 10−3 CL=90% 105
Hadronic modes with a K or K K K
K0
S
π+ ( 1.47±0.07) % S=2.0 863
K0
L
π+ ( 1.46±0.05) % 863
K− 2π+ [c] ( 9.13±0.19) % 846
(K− π+)S−wave π+ ( 7.32±0.19) % 846
K∗
0
(1430)0 π+ ,
K∗
0
(1430)0 → K− π+
[d] ( 1.21±0.06) % 382
HTTP://PDG.LBL.GOV Page 3 Created: 8/21/2014 13:13
D+
D-
K-
+
+
K+
-
-
p p
100 m
pBeam = 6.5 GeV/c

View Slide

D+
D-
K-
+
+
K+
-
-
→
R
p p
100 m
2015-01-15 19:08:40
R| / mm
∆
|
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
counts
10
2
10
3
10
4
10
5
10 Entries 1000000
Mean 0.5471
RMS 0.5424
R|
∆
|
-
D
+
D
2015-01-15 19:08:42
/ mm
xy
R)
∆
(
0 0.05 0.1 0.15 0.2 0.25 0.3
counts
0
10000
20000
30000
40000
50000
Entries 1000000
Mean 0.04196
RMS 0.03302
xy
R)
∆
(
-
D
+
D
2015-01-15 19:08:41
/ mm
z
R)
∆
(
5
− 4
− 3
− 2
− 1
− 0 1 2 3 4 5
counts
0
10000
20000
30000
40000
50000
60000
70000
80000
Entries 1000000
Mean 0.001421
−
RMS 0.7685
z
R)
∆
(
-
D
+
D
D Decay Vertex Distances
30

View Slide

D+
D-
K-
+
+
K+
-
-
p p
100 m
CHAPTER 4. APPLICATION TO THE RECONSTRUCTION OF D+ ! K ⇡+⇡+
2015-01-15 19:08:22
)
4
/c
2
/ (GeV
(1+2)
+
π
-
K
2
M
0.5 1 1.5 2 2.5 3
)
4
/c
2
counts / 0.03 (GeV
0
10000
20000
30000
40000
50000
Entries 2000000
Daughters Invariant Mass
+
D
(a) m2(K ⇡+). Two entries per event, as the two final
state ⇡ are indistinguishable.
2015-01-15 19:08:23
)
4
/c
2
/ (GeV
(2)
+
π
(1)
+
π
2
M
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
)
4
/c
2
counts / 0.02 (GeV
0
2000
4000
6000
8000
10000
12000
14000
Entries 1000000
+
D
(b) m2(⇡+⇡+)
Figure 4.7: Squared masses of the two two-particle-combinations of the D+ daughter particles.
4.6. BACKGROUND STUDIES
Figure 4.5: Momenta Plots from CLEO-c. The corresponding plots of EvtGen’s implementation are Figure
4.6, Figure 4.7(a), and Figure 4.7(b).
NOTE: Ref!
)
4
/c
2
) / (GeV
)
2.5
3
Entries 1000000
350
400
450
Entries 1000000
+
π
+
π
-
K
→
+
Dalitz plot for D
CHAPTER 4. APPLICATION TO THE RECONSTRUCTION OF D+ ! K ⇡+⇡
2015-01-15 19:08:22
)
4
/c
2
/ (GeV
(1+2)
+
π
-
K
2
M
0.5 1 1.5 2 2.5 3
)
4
/c
2
counts / 0.03 (GeV
0
10000
20000
30000
40000
50000
Entries 2000000
+
D
(a) m2(K ⇡+). Two entries per event, as the two final
state ⇡ are indistinguishable.
2015-01-15 19:08:23
)
4
/c
2
/ (GeV
(2)
+
π
(1)
+
π
2
M
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
)
4
/c
2
counts / 0.02 (GeV
0
2000
4000
6000
8000
10000
12000
14000
Entries 1000000
+
D
(b) m2(⇡+⇡+)
Figure 4.7: Squared masses of the two two-particle-combinations of the D+ daughter particles.
Decay Model: EvtGen / CLEO
31
CLEO-c
EvtGen
4.6. BACKGROUND STUDIES
Figure 4.5: Momenta Plots from CLEO-c. The corresponding plots of EvtGen’s implementation are Figure
4.6, Figure 4.7(a), and Figure 4.7(b).
NOTE: Ref!
2014-12-21 20:29:11
)
4
/c
2
) / (GeV
(1)
+
π
-
(K
2
m
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
)
4
/c
2
) / (GeV
(2)
+
π
-
(K
2
m
1
1.5
2
2.5
3
Entries 1000000
0
50
100
150
200
250
300
350
400
450
Entries 1000000
+
π
+
π
-
K
→
+
Dalitz plot for D
Figure 4.6: Dalitz plot for D+ ! K ⇡+⇡+. Since the two ⇡ in the final state are indistinguishable, the
distribution is symmetric around the diagonal y = x. K ⇡+
(1)
is the lighter combination, plotted on
x.

View Slide

2015-01-07 14:34:50
/ GeV/c
z
p
0 0.5 1 1.5 2 2.5 3 3.5
/ GeV/c
t
p
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1 Entries 2000000
0
200
400
600
800
1000
1200
1400
Entries 2000000
Momentum Distribution
(1+2)
+
π
K, π Momentum Distributions
32
2015-01-07 14:34:42
/ GeV/c
z
p
0 0.5 1 1.5 2 2.5 3 3.5
/ GeV/c
t
p
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1 Entries 1000000
0
100
200
300
400
500
Entries 1000000
Momentum Distribution
-
K

View Slide

D± → K∓ π± π±
33
Detector Response

View Slide

2015-01-27 14:49:51
#Hits
0 10 20 30 40 50 60 70 80
counts
0
10000
20000
30000
40000
50000
60000
Entries 200000
Mean 0.02411
±
12.23
RMS 0.01705
±
10.78
Underflow 0
Overflow 0
: #STT Hits
-
K
2015-01-28 20:29:58
#Hits
0 10 20 30 40 50 60
counts
1
10
2
10
3
10
4
10
5
10
Entries 200000
Mean 0.02405
±
3.443
RMS 0.01701
±
10.76
Underflow 0
Overflow 0
: #FTS Hits
-
K
Sub-Detector Hit Counting
34
2015-01-24 16:10:55
#Hits
0 1 2 3 4 5 6 7 8
counts
0
20
40
60
80
100
3
10
×
Entries 200000
Mean 0.006149
±
2.342
RMS 0.004348
±
2.75
Underflow 0
Overflow 0
: #GEM Hits
-
K
2015-01-24 16:10:54
#Hits
0 1 2 3 4 5 6 7 8 9
counts
0
10000
20000
30000
40000
50000
60000
Entries 200000
Mean 0.003445
±
3.671
RMS 0.002436
±
1.541
Underflow 0
Overflow 0
: #MVD Hits
-
K
K-
MVD STT
GEM FTS

View Slide

2015-01-28 20:39:59
#GEM Hits
0 1 2 3 4 5 6 7 8
#STT Hits
0
5
10
15
20
25
30
Entries 200000
Mean x 0.006153
±
2.343
Mean y 0.02403
±
12.2
RMS x 0.004351
±
2.751
RMS y 0.01699
±
10.74
0 185 0
0 199815 0
0 0 0
0
5000
10000
15000
20000
25000
30000
Entries 200000
Mean x 0.006153
±
2.343
Mean y 0.02403
±
12.2
RMS x 0.004351
±
2.751
RMS y 0.01699
±
10.74
0 185 0
0 199815 0
0 0 0
: #GEM Hits vs. #STT Hits
-
K
35
K-
STT
GEM

View Slide

2015-01-24 16:11:25
#Hits
0 1 2 3 4 5 6 7 8
counts
0
50
100
150
200
250
3
10
×
Entries 400000
Mean 0.003999
±
1.647
RMS 0.002828
±
2.529
Underflow 0
Overflow 8
: #GEM Hits
+
π
2015-01-24 16:11:24
#Hits
0 1 2 3 4 5 6 7 8 9
counts
0
20
40
60
80
100
120
3
10
×
Entries 400000
Mean 0.002305
±
3.815
RMS 0.00163
±
1.458
Underflow 0
Overflow 41
: #MVD Hits
+
π
36
π+
2015-01-28 20:30:27
#Hits
0 10 20 30 40 50 60 70 80
counts
0
10000
20000
30000
40000
50000
60000
70000
Entries 400000
Mean 0.01823
±
17.59
RMS 0.01289
±
11.51
Underflow 0
Overflow 1132
: #STT Hits
+
π
2015-01-28 20:33:00
#Hits
0 10 20 30 40 50 60
counts
1
10
2
10
3
10
4
10
5
10
Entries 400000
Mean 0.0152
±
2.582
RMS 0.01075
±
9.612
Underflow 0
Overflow 12
: #FTS Hits
+
π
MVD STT
GEM FTS

View Slide

2015-01-25 17:09:12
/ GeV/c
z
p
0 0.5 1 1.5 2 2.5 3 3.5
/ GeV/c
t
p
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Entries 70379
Mean x 0.003239
±
1.435
Mean y 0.0007536
±
0.3081
RMS x 0.00229
±
0.8592
RMS y 0.0005329
±
0.1999
0 0 0
0 70379 0
0 0 0
0
20
40
60
80
100
120
Entries 70379
Mean x 0.003239
±
1.435
Mean y 0.0007536
±
0.3081
RMS x 0.00229
±
0.8592
RMS y 0.0005329
±
0.1999
0 0 0
0 70379 0
0 0 0
for no #STT Hits
z
vs. p
t
: p
+
π
2015-01-25 17:09:13
/ GeV/c
z
p
0 0.5 1 1.5 2 2.5 3 3.5
/ GeV/c
t
p
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Entries 34085
Mean x 0.004096
±
0.962
Mean y 0.001204
±
0.3999
RMS x 0.002896
±
0.7562
RMS y 0.0008514
±
0.2223
0 0 0
0 34085 0
0 0 0
0
20
40
60
80
100
Entries 34085
Mean x 0.004096
±
0.962
Mean y 0.001204
±
0.3999
RMS x 0.002896
±
0.7562
RMS y 0.0008514
±
0.2223
0 0 0
0 34085 0
0 0 0
for no #STT, #GEM, #FTS Hits
z
vs. p
t
: p
+
π
2015-01-25 17:09:13
#FTS Hits
0 10 20 30 40 50 60
#GEM Hits
0
1
2
3
4
5
6
7
8
Entries 70379
Mean x 0.06796
±
12.04
Mean y 0.01082
±
2.685
RMS x 0.04805
±
18.03
RMS y 0.007649
±
2.869
0 1 0
0 70376 2
0 0 0
0
5000
10000
15000
20000
25000
30000
Entries 70379
Mean x 0.06796
±
12.04
Mean y 0.01082
±
2.685
RMS x 0.04805
±
18.03
RMS y 0.007649
±
2.869
0 1 0
0 70376 2
0 0 0
: #Hits FTS vs. #Hits GEM for no #Hits STT
+
π
37
π+: No STT hits
2015-01-28 20:30:27
#Hits
0 10 20 30 40 50 60 70 80
counts
0
10000
20000
30000
40000
50000
60000
70000
Entries 400000
Mean 0.01823
±
17.59
RMS 0.01289
±
11.51
Underflow 0
Overflow 1132
: #STT Hits
+
π
STT
GEM
FTS
STT
STT STT
GEM
FTS
MVD

View Slide

2015-01-30 11:41:07
θ
0 0.5 1 1.5 2 2.5 3
#Hits per track
0
5
10
15
20
25
30
35
40
Sub-Detectors
MVD
STT
GEM
FTS
ALL
for Different Sub-Detectors
θ
: Profile of #Hits vs.
+
π
2015-01-30 11:40:42
θ
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
#Hits per track
0
5
10
15
20
25
30 Sub-Detectors
MVD
STT
GEM
FTS
ALL
for Different Sub-Detectors
θ
: Profile of #Hits vs.
-
K
38
π & K: Profile histogram comparison

View Slide

2015-02-01 20:24:34
θ
0 0.5 1 1.5 2 2.5 3
#Hits > 6 / #Hits
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
(All Sub-Detectors)
θ
: Efficiency vs.
+
π
2015-02-01 20:24:03
θ
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
#Hits > 6 / #Hits
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
(All Sub-Detectors)
θ
: Efficiency vs.
-
K
39
π: Eﬃciency estimation
Tracks with more than 6 hits
All tracks
All All

View Slide

Decay Positions
40
K
2015-02-01 20:37:11
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
1
10
2
10
3
10
Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
K: Decay Positions r vs. z
All

View Slide

2015-02-01 20:37:11
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
1
10
2
10
3
10
Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
Decay Positions
41
2015-02-01 20:37:17
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
1
10
2
10
3
10
Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
: Decay Positions r vs. z
π
K & π
All
All

View Slide

2015-02-01 20:37:11
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
1
10
2
10
3
10
Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
Decay Positions
41
2015-02-01 20:37:17
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
1
10
2
10
3
10
Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
π
2015-02-01 20:37:13
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
1
10
2
10
3
10
Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
2015-02-01 20:37:18
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
1
10
2
10
3
10
Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
π
K & π
All
All

View Slide

2015-02-01 20:37:11
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
1
10
2
10
3
10
Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
Decay Positions
41
2015-02-01 20:37:17
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
1
10
2
10
3
10
Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
π
2015-02-01 20:37:13
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
1
10
2
10
3
10
Entries 382674
Mean x 0.306
±
205.9
Mean y 0.05362
±
54.9
RMS x 0.2164
±
189.3
RMS y 0.03791
±
33.16
0 66 4
0 382551 53
0 0 0
2015-02-01 20:37:18
z / cm
0 200 400 600 800 1000
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
1
10
2
10
3
10
Entries 762146
Mean x 0.203
±
161.2
Mean y 0.03335
±
60.81
RMS x 0.1436
±
177.2
RMS y 0.02358
±
29.11
0 59 3
163 761890 31
0 0 0
π
K & π
2015-02-01 20:37:16
θ
0 0.5 1 1.5 2 2.5 3
ratio of counts inside / all
0
0.2
0.4
0.6
0.8
1
K: Ratio of IN / ALL of DIRC/EMC
2015-02-01 20:37:22
θ
0 0.5 1 1.5 2 2.5 3
ratio of counts inside / all
0.2
0.4
0.6
0.8
1
: Ratio of IN / ALL of DIRC/EMC
π
All
All

View Slide

2015-01-25 18:00:32
z / cm
100
− 0 100 200 300 400 500 600
r / cm
0
20
40
60
80
100
120
140
160
180
200
220 Entries 33385
Mean x 0.3767
±
38.3
Mean y 0.1691
±
17.86
RMS x 0.2664
±
68.8
RMS y 0.1196
±
30.88
0 12 0
16 33354 3
0 0 0
0
200
400
600
800
1000
1200
1400
Entries 33385
Mean x 0.3767
±
38.3
Mean y 0.1691
±
17.86
RMS x 0.2664
±
68.8
RMS y 0.1196
±
30.88
0 12 0
16 33354 3
0 0 0
: Decay Positions r vs. z for no #STT, #GEM, #FTS Hits
+
π
Decay Positions
42
π+: Positions for no STT, no GEM, no FTS hits
Particles go through
uncovered slits!
STT
GEM
FTS
MVD

View Slide

D± → K∓ π± π±
43
Detector Response

View Slide

2015-02-02 14:42:50
θ
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
ε
0
0.2
0.4
0.6
0.8
1
Particles
All
±
K
±
π
θ
vs.
ε
Reconstruction efficiency
2015-02-02 14:46:07
θ
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
ε
0
0.2
0.4
0.6
0.8
1
Particles
All
±
K
±
π
θ
vs.
ε
Reconstruction efficiency
Circle Hough Benchmark
44
εAll = 97 %
εK = 82 %
επ = 95 %
K, π: Eﬃciency of reconstructable tracks
εAll = 81 %
εK = 71 %
επ = 82 %
PANDA
default
tracking
Circle
Hough

View Slide

2015-01-28 19:29:48
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
50
100
150
200
250
300 Entries 2076
Mean 0.002905
±
0.01168
−
RMS 0.002054
±
0.1323
/ ndf
2
χ 306.5 / 64
Constant 8.6
±
213.9
Mean 0.001599
±
0.008731
−
Sigma 0.00213
±
0.06589
Particles
±
Transverse Momentum Relative: #K
2015-01-28 19:29:46
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
100
200
300
400
500
600
700
800
900 Entries 5268
Mean 0.001937
±
0.02164
−
RMS 0.001369
±
0.14
/ ndf
2
χ 865 / 83
Constant 17.0
±
682
Mean 0.00078
±
0.01069
−
Sigma 0.00101
±
0.05104
Particles
±
π
Transverse Momentum Relative:
2015-01-28 19:27:47
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
100
200
300
400
500
600
Entries 2396
Mean 0.004085
±
0.04415
−
RMS 0.002889
±
0.1973
/ ndf
2
χ 526.1 / 87
Constant 21.0
±
611.8
Mean 0.000558
±
0.004154
−
Sigma 0.00059
±
0.02356
Particles
±
Transverse Momentum Relative: #K
2015-01-28 19:27:46
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
200
400
600
800
1000
1200
1400
1600
1800 Entries 6118
Mean 0.002647
±
0.03134
−
RMS 0.001872
±
0.2032
/ ndf
2
χ 1447 / 96
Constant 39.8
±
1846
Mean 0.000288
±
0.001895
−
Sigma 0.00030
±
0.01922
Particles
±
π
Transverse Momentum Relative:
45
K, π: pt resolution
σAll = 5.5 %
σK = 6.5 % σπ = 5.1 %
σAll = 2.1 %
σK = 2.3 % σπ = 1.9 %
σAll = 5.4 %
σK = 5.6 % σπ = 5.2 %
PANDA
default
tracking
Circle
Hough
PANDA
tracking
w/o Kalman

View Slide

Spurious found tracks:
Partially found tracks
46
pt resolution of diﬀerently well-found tracks
Fully found tracks
• All hits of reconstructed track
come from one MC track
• ≥ 50 % of hits of MC track are in
reconstructed track
• ≥ 50 % hits of reconstructed
track come from one MC track
• All hits of reconstructed track
come from one MC track
• All hits of MC track are in
reconstructed track
2015-01-28 19:29:49
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
100
200
300
400
500
600 Entries 3153
Mean 0.001133
±
0.005316
−
RMS 0.000801
±
0.06361
/ ndf
2
χ 261.2 / 33
Constant 14.1
±
487.3
Mean 0.000882
±
0.008452
−
Sigma 0.00105
±
0.04735
Transverse Momentum Relative: All, fully
2015-01-28 19:29:49
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
10
20
30
40
50
Entries 617
Mean 0.006823
±
0.002804
−
RMS 0.004825
±
0.1691
/ ndf
2
χ 95.64 / 50
Constant 2.26
±
34.85
Mean 0.00568
±
0.01461
−
Sigma 0.0057
±
0.1187
Transverse Momentum Relative: All, Partially
2015-01-28 19:29:50
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
50
100
150
200
250
300
350
400
450
Entries 2807
Mean 0.002806
±
0.01998
−
RMS 0.001984
±
0.1478
/ ndf
2
χ 497.5 / 79
Constant 11.8
±
351.8
Mean 0.00108
±
0.01299
−
Sigma 0.00134
±
0.05168
Transverse Momentum Relative: All, Spurious (>0.7)
7 %
2.1 %
13.7 %
PANDA
default
tracking
Circle
Hough
PANDA
tracking
w/o Kalman
σpt
N
11.9 %
11 %
11 %
41 %
5.1 %
7.4 %
σpt
N
5.1 %
36 %
36 %
34 %
1.8 %
4.1 %
σpt
N
4.7 %
47 %
47 %

View Slide

47
2015-01-28 19:29:50
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
50
100
150
200
250
300
350
400
450
Entries 2807
Mean 0.002806
±
0.01998
−
RMS 0.001984
±
0.1478
/ ndf
2
χ 497.5 / 79
Constant 11.8
±
351.8
Mean 0.00108
±
0.01299
−
Sigma 0.00134
±
0.05168
PANDA
default
tracking
Circle
Hough
PANDA
tracking
w/o Kalman
41 %
5.1 %
7.4 %
σpt
N
5.1 %
36 %
36 %
2015-01-28 19:29:50
fraction (a.u.)
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
counts
0
20
40
60
80
100
120
140
160
180 Entries 3821
Mean 0.002095
±
0.7873
RMS 0.001481
±
0.1295
Fraction of Hits of a MC Track in Assoc. Reco Track

View Slide

47
2015-01-28 19:29:50
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
50
100
150
200
250
300
350
400
450
Entries 2807
Mean 0.002806
±
0.01998
−
RMS 0.001984
±
0.1478
/ ndf
2
χ 497.5 / 79
Constant 11.8
±
351.8
Mean 0.00108
±
0.01299
−
Sigma 0.00134
±
0.05168
PANDA
default
tracking
Circle
Hough
PANDA
tracking
w/o Kalman
41 %
5.1 %
7.4 %
σpt
N
5.1 %
36 %
36 %
2015-01-28 19:29:50
fraction (a.u.)
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
counts
0
20
40
60
80
100
120
140
160
180 Entries 3821
Mean 0.002095
±
0.7873
RMS 0.001481
±
0.1295
2015-01-28 19:29:51
fraction reco'd hits (a.u.)
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
MC
t
) / p
MC
t
- p
RECO
t
(p
1
−
0.8
−
0.6
−
0.4
−
0.2
−
0
0.2
0.4
0.6
0.8
1
Entries 3821
Mean x 0.002086
±
0.7925
Mean y 0.003193
±
0.03289
RMS x 0.001475
±
0.1269
RMS y 0.002258
±
0.1942
0
5
10
15
20
25
30
35
40
45
Entries 3821
Mean x 0.002086
±
0.7925
Mean y 0.003193
±
0.03289
RMS x 0.001475
±
0.1269
RMS y 0.002258
±
0.1942
res.
t
Fraction of Hits of a MC Track in Assoc. Reco Track vs. p

View Slide

47
2015-01-28 19:29:50
MC
t
) / p
MC
t
- p
RECO
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
50
100
150
200
250
300
350
400
450
Entries 2807
Mean 0.002806
±
0.01998
−
RMS 0.001984
±
0.1478
/ ndf
2
χ 497.5 / 79
Constant 11.8
±
351.8
Mean 0.00108
±
0.01299
−
Sigma 0.00134
±
0.05168
PANDA
default
tracking
Circle
Hough
PANDA
tracking
w/o Kalman
41 %
5.1 %
7.4 %
σpt
N
5.1 %
36 %
36 %
2015-01-28 19:29:50
fraction (a.u.)
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
counts
0
20
40
60
80
100
120
140
160
180 Entries 3821
Mean 0.002095
±
0.7873
RMS 0.001481
±
0.1295
2015-01-28 19:29:51
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
MC
t
) / p
MC
t
- p
RECO
t
(p
1
−
0.8
−
0.6
−
0.4
−
0.2
−
0
0.2
0.4
0.6
0.8
1
Entries 3821
Mean x 0.002086
±
0.7925
Mean y 0.003193
±
0.03289
RMS x 0.001475
±
0.1269
RMS y 0.002258
±
0.1942
0
5
10
15
20
25
30
35
40
45
Entries 3821
Mean x 0.002086
±
0.7925
Mean y 0.003193
±
0.03289
RMS x 0.001475
±
0.1269
RMS y 0.002258
±
0.1942
res.
t
2015-02-02 16:06:04
0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1
MC
t
) / p
MC
t
- p
RECO
t
(p
0.1
−
0.05
−
0
0.05
0.1
0.15
0.2
0.25
0.3 Entries 3699
Mean 0.002084
±
0.7925
Mean y 0.003193
±
0.03276
RMS 0.001474
±
0.1267
RMS y 0.002258
±
0.1942
res.
t

View Slide

2015-02-02 14:36:57
/ GeV/c
MC
t
p
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ratio
ε
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Algorithms
Default
Circle Hough
Efficiency of All MC Tracks vs. Transv. Mom.
2015-02-02 14:36:58
/ GeV/c
MC
t
p
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ratio
ε
0
0.2
0.4
0.6
0.8
1
Algorithms
Default
Circle Hough
Efficiency of MC Tracks Possible to Find vs. Transv. Mom.
48
Eﬃciencies vs. pt
PANDA
default
tracking
Circle
Hough
Found tracks
All tracks
Found tracks
Findable tracks

View Slide

D± → K∓ π± π±
49
Detector Response

View Slide

Oﬄine Event Analysis
• PandaRoot: scrut14
• Beam Momentum: pBeam = 6.5 GeV/c
• PID algorithm: PidAlgoIdealCharged
• PID selection: {Kaon,Pi}Best{Plus,Minus}
• Initial signal dataset: 200 000 events produced
• Mass window cut: Δmcut = 0.15 GeV/c2 around mD
50

View Slide

2015-02-01 16:42:18
Mesons
-
#D
0 1 2 3 4 5 6
ratio (a.u.)
0
0.05
0.1
0.15
0.2
0.25
0.3 Entries 83512
/ event
-
# D
/ event
+
# D
D: Percentage of Reconstructed Mesons (before cuts)
Oﬄine Analysis — D Mesons / Event
51
• 25 % (30 %) of all events: 1 D+(-) is found
• 9 % (12 %) of all events: >1 D+(-) is found
no cuts

View Slide

Oﬄine Analysis — Daughter Momenta
52
2015-02-01 16:42:19
/ GeV/c
z
p
0 0.5 1 1.5 2 2.5 3 3.5
/ GeV/c
t
p
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Entries 69987
Mean x 0.002427
±
1.236
Mean y 0.0008045
±
0.4365
RMS x 0.001716
±
0.6412
RMS y 0.0005689
±
0.2126
1 85 37
43 69809 12
0 0 0
0
5
10
15
20
25
30
35
40
45
Entries 69987
Mean x 0.002427
±
1.236
Mean y 0.0008045
±
0.4365
RMS x 0.001716
±
0.6412
RMS y 0.0005689
±
0.2126
1 85 37
43 69809 12
0 0 0
): Momentum Distribution
+
(from D
-
K
2015-02-01 16:42:20
/ GeV/c
z
p
0 0.5 1 1.5 2 2.5 3
/ GeV/c
t
p
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Entries 139974
Mean x 0.001737
±
0.9904
Mean y 0.0004872
±
0.4571
RMS x 0.001229
±
0.6485
RMS y 0.0003445
±
0.1818
0 308 137
105 139299 125
0 0 0
0
10
20
30
40
50
60
70
80
90
100
Entries 139974
Mean x 0.001737
±
0.9904
Mean y 0.0004872
±
0.4571
RMS x 0.001229
±
0.6485
RMS y 0.0003445
±
0.1818
0 308 137
105 139299 125
0 0 0
): Momentum Distribution (2 Entries per D)
+
(from D
+
Pi
no cuts

View Slide

Oﬄine Analysis — D Candidate Mass
53
no cuts
2015-02-01 17:42:20
2
m / GeV/c
1.7 1.75 1.8 1.85 1.9 1.95 2 2.05
counts
0
500
1000
1500
2000
2500
3000
3500
4000
Entries 69987
Mean 0.0001896
±
1.862
RMS 0.0001341
±
0.05016
Underflow 0
Overflow 0
: Invariant Mass
+
D

View Slide

Oﬄine Analysis — Vertex Fit
54
no cuts
2015-02-01 17:42:21
2
χ
0 1 2 3 4 5 6 7 8 9 10
counts
0
200
400
600
800
1000
1200
1400
Entries 69987
Mean 0.009761
±
2.521
RMS 0.006902
±
2.142
Underflow 744
Overflow 2.11e+04
2
χ
: Vertex Fit
+
D
2015-02-01 17:42:21
)
2
χ
Prob(
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
counts
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
22000
Entries 69987
Mean 0.001346
±
0.3888
RMS 0.0009516
±
0.3556
Underflow 0
Overflow 146
)
2
χ
: Vertex Fit Prob(
+
D
PndKinVtxFit

View Slide

Oﬄine Analysis — Vertex Fit
54
no cuts
2015-02-01 17:42:21
2
χ
0 1 2 3 4 5 6 7 8 9 10
counts
0
200
400
600
800
1000
1200
1400
Entries 69987
Mean 0.009761
±
2.521
RMS 0.006902
±
2.142
Underflow 744
Overflow 2.11e+04
2
χ
: Vertex Fit
+
D
2015-02-01 17:42:21
)
2
χ
Prob(
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
counts
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
22000
Entries 69987
Mean 0.001346
±
0.3888
RMS 0.0009516
±
0.3556
Underflow 0
Overflow 146
)
2
χ
: Vertex Fit Prob(
+
D
Require Prob(χ2) > 0.01
PndKinVtxFit

View Slide

Oﬄine Analysis — Vertex After Vertex Fit
55
Vertex Fit
2015-02-01 17:42:25
z / cm
0.1
− 0.05
− 0 0.05 0.1 0.15 0.2 0.25
counts
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400 Entries 69987
Mean 0.0002301
±
0.05153
RMS 0.0001627
±
0.05695
Underflow 5395
Overflow 3341
Entries 42207
Mean 0.0002513
±
0.05143
RMS 0.0001777
±
0.05111
Underflow 222
Overflow 605
: z Position After Vtx Fit (Comparison)
+
D
2015-02-01 17:42:23
x / cm
0.1
− 0.08
− 0.06
− 0.04
− 0.02
− 0 0.02 0.04 0.06 0.08 0.1
counts
0
1000
2000
3000
4000
5000
6000
Entries 69987
Mean 05
−
8.289e
±
06
−
3.428e
RMS 05
−
5.861e
±
0.02081
Underflow 3384
Overflow 3590
Entries 42207
Mean 05
−
5.393e
±
05
−
1.866e
−
RMS 05
−
3.814e
±
0.01106
Underflow 87
Overflow 96
: x Position After Vtx Fit (Comparison)
+
D
2015-02-01 17:42:24
y / cm
0.1
− 0.08
− 0.06
− 0.04
− 0.02
− 0 0.02 0.04 0.06 0.08 0.1
counts
0
1000
2000
3000
4000
5000
6000
Entries 69987
Mean 05
−
8.361e
±
05
−
1.227e
RMS 05
−
5.912e
±
0.02101
Underflow 3434
Overflow 3395
Entries 42207
Mean 05
−
5.304e
±
06
−
4.617e
RMS 05
−
3.75e
±
0.01087
Underflow 99
Overflow 110
: y Position After Vtx Fit (Comparison)
+
D
no cuts

View Slide

Oﬄine Analysis — Mass After Vertex Fit
56
no cuts
Vertex Fit
2015-02-01 18:13:26
2
) / GeV/c
MC
(m - m
0.15
− 0.1
− 0.05
− 0 0.05 0.1 0.15
counts
0
500
1000
1500
2000
2500
3000
3500
4000
Entries 69987
Mean 0.0001858
±
0.007714
−
RMS 0.0001314
±
0.04639
Underflow 1468
Overflow 6176
/ ndf
2
χ 1.241e+04 / 97
Constant 24.8
±
3255
Mean 0.000100
±
0.002007
−
Sigma 0.00014
±
0.02204
Entries 42207
Mean 0.0001669
±
0.003508
−
RMS 0.000118
±
0.03381
Underflow 98
Overflow 1055
/ ndf
2
χ 4654 / 94
Constant 22.4
±
2882
Mean 0.000098
±
0.001167
−
Sigma 0.00010
±
0.01814
: Relative Invariant Mass After Vertex Fit Comparison
+
D
2015-02-01 18:16:44
) / GeV/c
MC
t
- p
t
(p
1
− 0.8
− 0.6
− 0.4
− 0.2
− 0 0.2 0.4 0.6 0.8 1
counts
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
22000 Entries 69987
Mean 0.0005834
±
0.0542
RMS 0.0004125
±
0.1531
Underflow 0
Overflow 1106
/ ndf
2
χ 1.764e+04 / 55
Constant 1.520e+02
±
2.379e+04
Mean 0.000077
±
0.001102
Sigma 0.00008
±
0.01719
Entries 42207
Mean 0.0009259
±
0.01594
RMS 0.0006547
±
0.1873
Underflow 0
Overflow 1284
/ ndf
2
χ 5474 / 97
Constant 23.3
±
2732
Mean 0.000551
±
0.003463
Sigma 0.0007
±
0.1035
: Relative Transverse Momentum After Vertex Fit Comparison
+
D

View Slide

2015-02-01 18:16:45
2
χ
0 1 2 3 4 5 6 7 8 9 10
counts
0
2000
4000
6000
8000
10000
12000
Entries 69987
Mean 0.008903
±
1.501
RMS 0.006295
±
2.094
Underflow 219
Overflow 1.443e+04
2
χ
: Mass Fit
+
D
2015-02-01 18:16:45
)
2
χ
Prob(
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
counts
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
Entries 69987
Mean 0.001257
±
0.3587
RMS 0.0008885
±
0.3321
Underflow 0
Overflow 142
)
2
χ
: Mass Fit Prob(
+
D
Oﬄine Analysis — Mass Constraint Fit
57
no cuts
PndKinFitter with D+ mass constraint

View Slide

2015-02-01 18:16:45
2
χ
0 1 2 3 4 5 6 7 8 9 10
counts
0
2000
4000
6000
8000
10000
12000
Entries 69987
Mean 0.008903
±
1.501
RMS 0.006295
±
2.094
Underflow 219
Overflow 1.443e+04
2
χ
: Mass Fit
+
D
2015-02-01 18:16:45
)
2
χ
Prob(
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
counts
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
Entries 69987
Mean 0.001257
±
0.3587
RMS 0.0008885
±
0.3321
Underflow 0
Overflow 142
)
2
χ
: Mass Fit Prob(
+
D
57
no cuts
Require Prob(χ2) > 0.01
PndKinFitter with D+ mass constraint

View Slide

2015-02-01 18:35:24
2
) / GeV/c
MC
(m - m
0.15
− 0.1
− 0.05
− 0 0.05 0.1 0.15
counts
0
500
1000
1500
2000
2500
3000
3500
4000
4500 Entries 69987
Mean 0.0001816
±
0.007284
−
RMS 0.0001284
±
0.04597
Underflow 0
Overflow 5940
/ ndf
2
χ 1.312e+04 / 81
Constant 25.5
±
3265
Mean 0.000101
±
0.002073
−
Sigma 0.0001
±
0.0224
Entries 42207
Mean 0.0001669
±
0.003508
−
RMS 0.000118
±
0.03381
Underflow 98
Overflow 1055
/ ndf
2
χ 4654 / 94
Constant 22.4
±
2882
Mean 0.000098
±
0.001167
−
Sigma 0.00010
±
0.01814
Entries 45375
Mean 0.0001155
±
0.001035
−
RMS 05
−
8.169e
±
0.024
Underflow 117
Overflow 2095
/ ndf
2
χ 1609 / 93
Constant 22.1
±
3388
Mean 0.0000866
±
0.0009225
−
Sigma 0.00008
±
0.01762
Entries 35182
Mean 0.0001144
±
0.0006527
−
RMS 05
−
8.09e
±
0.02126
Underflow 50
Overflow 583
/ ndf
2
χ 969.4 / 90
Constant 21.3
±
2940
Mean 0.0000897
±
0.0007204
−
Sigma 0.00008
±
0.01641
: Relative Mass After Fits Comparison
+
D
58
• In 17.5 % (20.5 %) of all events: D+ (D-) is found
• In 8.6 % of all events: D+D- are found
• DPM: Of 1M events, 100 survive these fits
→ more cuts!
no cuts
Vertex Fit
Mass Fit
2015-02-02 11:24:43
2
) / GeV/c
MC
(m - m
0.15
− 0.1
− 0.05
− 0 0.05 0.1 0.15
counts
0
500
1000
1500
2000
2500
3000
3500
4000
4500 Entries 69987
Mean 0.0001816
±
0.007284
−
RMS 0.0001284
±
0.04597
Underflow 0
Overflow 5940
/ ndf
2
χ 1.312e+04 / 81
Constant 25.5
±
3265
Mean 0.000101
±
0.002073
−
Sigma 0.0001
±
0.0224
Entries 42207
Mean 0.0001669
±
0.003508
−
RMS 0.000118
±
0.03381
Underflow 98
Overflow 1055
/ ndf
2
χ 4654 / 94
Constant 22.4
±
2882
Mean 0.000098
±
0.001167
−
Sigma 0.00010
±
0.01814
Entries 45375
Mean 0.0001155
±
0.001035
−
RMS 05
−
8.169e
±
0.024
Underflow 117
Overflow 2095
/ ndf
2
χ 1609 / 93
Constant 22.1
±
3388
Mean 0.0000866
±
0.0009225
−
Sigma 0.00008
±
0.01762
Entries 35182
Mean 0.0001144
±
0.0006527
−
RMS 05
−
8.09e
±
0.02126
Underflow 50
Overflow 583
/ ndf
2
χ 969.4 / 90
Constant 21.3
±
2940
Mean 0.0000897
±
0.0007204
−
Sigma 0.00008
±
0.01641
: Relative Mass After Fits Comparison
+
D
Mass & Vtx

View Slide

Oﬄine Analysis — Inclusive Event
59
• When a D+ is found, the invariant missing mass is at:
µ = (1.870 ± 0.0009) GeV/c2 ,
σ = (0.01676 ± 0.00009) GeV/c2
• PDG: 1.869 GeV/c2
no cuts
Vertex Fit
2015-02-02 11:24:48
2
m / GeV/c
1.7 1.75 1.8 1.85 1.9 1.95 2 2.05
counts
0
500
1000
1500
2000
2500
3000
3500
4000
Entries 69987
Mean 0.0001811
±
1.871
RMS 0.0001281
±
0.04629
Underflow 4485
Overflow 169
/ ndf
2
χ 1.262e+04 / 93
Constant 24.4
±
3066
Mean 0.000
±
1.872
Sigma 0.000
±
0.024
Entries 42207
Mean 0.0001673
±
1.871
RMS 0.0001183
±
0.03388
Underflow 1086
Overflow 117
/ ndf
2
χ 4516 / 96
Constant 21.4
±
2741
Mean 0.000
±
1.871
Sigma 0.00011
±
0.01858
Entries 35182
Mean 0.0001217
±
1.869
RMS 05
−
8.606e
±
0.02265
Underflow 472
Overflow 68
/ ndf
2
χ 1159 / 96
Constant 20.5
±
2790
Mean 0.00
±
1.87
Sigma 0.00008
±
0.01676
: Inclusive Missing Mass After Fits Comparison
+
D
Mass & Vtx

View Slide

Summary
• GPU Algorithms:
– Diﬀerent algorithms in development
– Triplet Finder: Max performance 12 MHit/s
– Circle Hough: ≥ 70 % reconstruction performance
• Benchmark channel D → K π π
– 17.5 % D+ can be reconstructed (37 % before fits)
~1300 D+ / h (σ = 100 nb)
→ Continue studies (Background, Circle Hough)
60

View Slide

Thank you!
Andreas Herten
[email protected]
Summary
• GPU Algorithms:
– Diﬀerent algorithms in development
– Triplet Finder: Max performance 12 MHit/s
– Circle Hough: ≥ 70 % reconstruction performance
• Benchmark channel D → K π π
– 17.5 % D+ can be reconstructed (37 % before fits)
~1300 D+ / h (σ = 100 nb)
→ Continue studies (Background, Circle Hough)
60

View Slide

BACKUP
61

View Slide

STT — Drift Tubes and t0
62

View Slide

62
Particle ionizes gas atoms in drift tubes

View Slide

62
Electrons drift to anode wire, ions to wall

View Slide

62
Signal only when electrons arrive at wire
No information about drift duration!
For that, start time (t0) needed:
t0 - tarrival ≈ tdrift
vdrift = const → tdrift • vdrift = risochrone

View Slide

62
risochrone

View Slide

62
Usual HEP experiment: t0 by trigger
But PANDA has no trigger…
risochrone

View Slide

Triplet Finder — Method
63
STT
More

View Slide

• STT hit in pivot straw
63
STT
More

View Slide

• Find surrounding hits
→ Create virtual hit (triplet)
at center of gravity (cog)
63
STT
More

View Slide

• Combine with
63
STT
More

View Slide

• Combine with
1.Second STT pivot-cog virtual hit
63
STT
More

View Slide

• Combine with
2.Interaction point
63
Interaction Point
STT
More

View Slide

• Combine with
2.Interaction point
• Calculate circle through three
points
63
Interaction Point
STT
More

View Slide

• Combine with
2.Interaction point
• Calculate circle through three
points
→ Track Candidate
63
Interaction Point
STT
More

View Slide

Triplet Finder — Animation
64
Triplet
Isochrone early
Isochrone early & skewed
Isochrone close
Isochrone late
MVD hit
Track timed out
Track current

View Slide

Dynamic
Parallelism
• Compare data processing strategies
65
1 thread/bunch
Calling
kernel
1 thread/bunch
Calling
kernel
Triplet
Finder
1 thread/bunch
Calling
kernel
1 block/bunch
Joined
kernel
1 block/bunch
Joined
kernel
1 block/bunch
Joined
kernel
TF Stage #1
TF Stage #2
TF Stage #3
TF Stage #4
1 stream/bunch
Combining
stream
1 stream/bunch
Combining
stream
1 stream/bunch
Calling
stream
Joined
Kernel
Host
Streams
Triplet
Finder
Triplet
Finder
CPU
GPU
TF Stage #1
TF Stage #2
TF Stage #3
TF Stage #4
TF Stage #1
TF Stage #2
TF Stage #3
TF Stage #4

View Slide

66
Triplet Finder — Data Processing
Explanation
K20X

View Slide

Triplet Finder — Binning: Sector Rows
67
• Sector Row testing
– After found track:
Hit association not with all hits of current window,
but only with subset
(first test rows of sector, then hits of row)
More

View Slide

68
Triplet Finder — Binning: Sector Rows
K20X
All Tubes (No Binning)
Sector-Row Binning

View Slide

69
Triplet Finder — Binning: Skewlets
K20X
Skewlet Binning
All Skewlets (No Binning)

View Slide

70
Triplet Finder — AoS vs. SoA
K20X
struct {
int r, g, b;
} AoS[N];
struct {
int r[N];
int g[N];
int b[N];
} SoA;

View Slide

71
Triplet Finder — CUDA Versions
K20X

View Slide

72
Triplet Finder — Clock Speed / GPU
K40 3004 MHz, 745 MHz / 875 MHz
K20X 2600 MHz, 732 MHz / 784 MHz
Memory Clock Core Clock GPU Boost

View Slide

73
Triplet Finder — Comparison to Kepler
K20X 750 Ti
Kepler
Price: 3600 €
Maxwell
Price: 130 €

View Slide

74
K20X 750 Ti

View Slide

75
Performance per multiprocessor
K20X
750 Ti

View Slide

Tesla K40 Tesla K20X GeForce GTX 760 Ti
Peak double
performance
Peak single
performance
GPU Chipset
# CUDA Cores
Memory size
Memory
bandwidth
1.46 TFLOPS 1.31 TFLOPS 0.04 TFLOPS
4.29 TFLOPS 3.95 TFLOPS 1.3 TFLOPS
GK110B GK110 GM107
2880 2688 640
12 GB 6 GB 2 GB
288 GByte/s 250 GByte/s 192 GByte/s
• Impact of chipset
76
Source: http://www.nvidia.com/content/tesla/pdf/NVIDIA-Tesla-Kepler-Family-Datasheet.pdf

View Slide

Triplet Finder — Kernel Launch Strategies
• Joined Kernel (JK): slowest
– High # registers → low occupancy
• Dynamic Parallelism (DP) / Host Streams (HS): comparable performance
– Performance
• HS faster for small # processed hits, DP faster for > 45000 hits
• HS stagnates there, while DP continues rising
– Limiting factor
• High # of required kernel calls
• Kernel launch latency
• Memcopy
– HS more aﬀected by this, because
• More PCI-E transfers (launch configurations for kernels)
• Less launch throughput, kernel launch latency gets more important
• False dependencies of launched kernels
– Single CPU thread handles all CUDA streams (Multi-thread possible, but
synchronization overhead too high for good performance)
– Grid scheduling done on hardware (Grid Management Unit) (DP: software)
» False dependencies when N(streams) > N(device connections)=323.5
77
Back

View Slide

78
ALGORITHMS #2
Hough Transform
Riemann Track Finder
Triplet Finder

View Slide

79
Riemann Track Finder — Method
• Idea: Don‘t fit lines (in 2D), fit planes (in 3D)!
• Create seeds
– All possible three hit combinations
• Grow seeds to tracks
Continuously test next hit if it fits
– Use mapping to Riemann paraboloid (+ s-z fit, det. layer)
x
x
x
x
y
z‘
x
x
x
y
x
x
x
x
y
x
More on: Seeds; Growing
1
2

View Slide

80
1 2 3 4 5
1
2
3
4
5
Riemann Algorithm — 1 Triplets
1
Layer number

View Slide

80
1 2 3 4 5
21
11 31
1
2
3
4
5
1
Layer number

View Slide

80
1 2 3 4 5
21
11 31
31
11 41
1
2
3
4
5
1
Layer number

View Slide

80
1 2 3 4 5
21
11 31
31
11 41
31
11 32
1
2
3
4
5
1
Layer number

View Slide

81
Riemann Algorithm — 1 Expansion
2

View Slide

81
2
x
x
x
x
y
z‘
Expand to z‘

View Slide

81
2
x
x
x
x
y
z‘
Expand to z‘
x
x
x
y
x
Riemann Surface
(paraboloid)

View Slide

81
2
x
x
x
x
y
z‘
Expand to z‘
x
x
x
y
x
Riemann Surface
(paraboloid)
x

View Slide

82
Riemann Track Finder — GPU Adaptations
CPU GPU

View Slide

82
CPU GPU
3 loops to generate seeds
serially
for (int i = 0; i < hitsInLayerOne.size(); i++) {
for (int j = 0; j < hitsInLayerTwo.size(); j++) {
for (int k = 0; k < hitsInLayerThree.size(); k++) {
/* Triplet Generation */
}
}
}
Needed: Mapping of
inherent GPU indexing
variable to triplet index
int ijk = threadIdx.x + blockIdx.x * blockDim.x;
nLayerx
= 1
2
⇣p
8x
+
1 1
⌘
pos
(
nLayerx
) =
3
pp
3
p
243x2 1
+
27x
32
/
3
+ 1
3
p
3
3
pp
3
p
243x2 1
+
27x
1
1

View Slide

82
CPU GPU
serially
}
}
}
Needed: Mapping of
nLayerx
= 1
2
⇣p
8x
+
1 1
⌘
pos
(
nLayerx
) =
3
pp
3
p
243x2 1
+
27x
32
/
3
+ 1
3
p
3
3
pp
3
p
243x2 1
+
27x
1
1
2
Port of CPU code;
parallelism on seed base
Only easy computations;
e.g. 3x3 matrices

View Slide

82
CPU GPU
→ 100 × faster than CPU version: ~0.6 ms/event
Still needs merging into PandaRoot
serially
}
}
}
Needed: Mapping of
nLayerx
= 1
2
⇣p
8x
+
1 1
⌘
pos
(
nLayerx
) =
3
pp
3
p
243x2 1
+
27x
32
/
3
+ 1
3
p
3
3
pp
3
p
243x2 1
+
27x
1
1
2
Port of CPU code;
parallelism on seed base
Only easy computations;
e.g. 3x3 matrices

View Slide

• Idea: Transform (x,y)i → (α,r)ij, find lines via (α,r) space
• Solve rij line equation for
– Lots of hits (x,y,ρ)i
and
– Many αj ∈ [0°,360°) each
• Fill histogram
• Extract track parameters
83
x
y
x
y
→ Bin
giv
r
α

View Slide

• Idea: Transform (x,y)i → (α,r)ij, find lines via (α,r) space
• Solve rij line equation for
– Lots of hits (x,y,ρ)i
and
– Many αj ∈ [0°,360°) each
• Fill histogram
• Extract track parameters
83
rij =
cos
↵j
·
xi +
sin
↵j
·
yi + ⇢i
i: ~100 hits/event (STT)
j: every 0.2° rij: 180 000
x
y
x
y
→ Bin
giv
r
α

View Slide

Hough Transform — Granularity
84
• Line Hough around point
α = 0°, 2°, 4°, …
α = 0°, 2°, 4°, …
rij = cos(αj) ⋅ xi + sin(αj) ⋅ yi static slide

View Slide

85
• Line Hough with isochrones
α = 0°, 10°, 20°, …
α = 0°, 10°, 20°, …
rij
± = cos(αj) ⋅ xi + sin(αj) ⋅ yi ± ρi static slide

View Slide

Hough Transform — Remarks
86
°
/
α
0 20 40 60 80 100 120 140 160 180
r
-30
-20
-10
0
10
20
30
40
HoughHist
Entries 9000
Mean x 89.33
Mean y 6.66
RMS x 51.8
RMS y 19.2
0
2
4
6
8
10
12
14
16
18
HoughHist
Entries 9000
Mean x 89.33
Mean y 6.66
RMS x 51.8
RMS y 19.2
HT histogram
Hill Climber
Peakfinding challenging

View Slide

Hough Transform — Remarks
86
°
/
α
0 20 40 60 80 100 120 140 160 180
r
-30
-20
-10
0
10
20
30
40
houghIt0
Entries 9000
Mean x 89.33
Mean y 6.66
RMS x 51.8
RMS y 19.2
0
2
4
6
8
10
12
14
16
18
houghIt0
Entries 9000
Mean x 89.33
Mean y 6.66
RMS x 51.8
RMS y 19.2
HT histogram
°
/
α
0 20 40 60 80 100 120 140 160 180
r
-30
-20
-10
0
10
20
30
40
houghIt1
Entries 5580
Mean x 89.6
Mean y 9.719
RMS x 51.78
RMS y 18.09
0
2
4
6
8
10
12
14
16
houghIt1
Entries 5580
Mean x 89.6
Mean y 9.719
RMS x 51.78
RMS y 18.09
HT histogram
°
/
α
0 20 40 60 80 100 120 140 160 180
r
-30
-20
-10
0
10
20
30
houghIt2
Entries 2700
Mean x 89.13
Mean y 13.79
RMS x 51.77
RMS y 14.04
0
2
4
6
8
10
12
houghIt2
Entries 2700
Mean x 89.13
Mean y 13.79
RMS x 51.77
RMS y 14.04
HT histogram
-40 -30 -20 -10 0 10 20 30 40
0
5
10
15
20
25
30
Iterative Maximum Deleter
Peakfinding challenging

View Slide

2015-01-24 16:11:25
#Hits
0 10 20 30 40 50 60
counts
0
50
100
150
200
250
300
350
3
10
×
Entries 400000
Mean 0.0152
±
2.582
RMS 0.01075
±
9.612
Underflow 0
Overflow 12
: #FTS Hits
+
π
2015-01-24 16:11:25
#Hits
0 1 2 3 4 5 6 7 8
counts
0
50
100
150
200
250
3
10
×
Entries 400000
Mean 0.003999
±
1.647
RMS 0.002828
±
2.529
Underflow 0
Overflow 8
: #GEM Hits
+
π
2015-01-24 16:11:27
θ
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
counts
0
100
200
300
400
500
600
700
800
Entries 29006
Mean 0.0003544
±
0.1155
RMS 0.0002506
±
0.05849
Underflow 0
Overflow 1768
for #FTS Hits > 0
θ
:
+
π
87
π+: θ for no GEM hits and for no FTS hits
2015-01-25 17:09:14
θ
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
counts
0
500
1000
1500
2000
2500
3000
3500
4000
Entries 131877
Mean 0.0003124
±
0.2456
RMS 0.0002209
±
0.1129
Underflow 0
Overflow 1378
for #GEM Hits > 0
θ
:
+
π

View Slide

2015-01-25 17:22:23
θ
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
#Hits
0
10
20
30
40
50
60
Entries 200000
Mean x 0.0004736
±
0.354
Mean y 0.0244
±
3.496
RMS x 0.0003349
±
0.2102
RMS y 0.01726
±
10.83
0 0 0
0 196952 3048
0 0 0
0
500
1000
1500
2000
2500
3000
3500
4000
Entries 200000
Mean x 0.0004736
±
0.354
Mean y 0.0244
±
3.496
RMS x 0.0003349
±
0.2102
RMS y 0.01726
±
10.83
0 0 0
0 196952 3048
0 0 0
θ
: #Hits FTS vs.
-
K
88
K-: #Hits vs. θ
2015-01-25 17:22:22
θ
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
#Hits
0
1
2
3
4
5
6
7
8
Entries 200000
Mean x 0.0004736
±
0.354
Mean y 0.006209
±
2.379
RMS x 0.0003349
±
0.2102
RMS y 0.004391
±
2.756
0 0 0
0 196952 3048
0 0 0
0
500
1000
1500
2000
2500
Entries 200000
Mean x 0.0004736
±
0.354
Mean y 0.006209
±
2.379
RMS x 0.0003349
±
0.2102
RMS y 0.004391
±
2.756
0 0 0
0 196952 3048
0 0 0
θ
: #Hits GEM vs.
-
K
2015-01-25 17:22:21
θ
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
#Hits
0
1
2
3
4
5
6
7
8
9 Entries 200000
Mean x 0.0004736
±
0.354
Mean y 0.003475
±
3.685
RMS x 0.0003349
±
0.2102
RMS y 0.002457
±
1.542
0 0 0
0 196952 3048
0 0 0
0
500
1000
1500
2000
2500
3000
Entries 200000
Mean x 0.0004736
±
0.354
Mean y 0.003475
±
3.685
RMS x 0.0003349
±
0.2102
RMS y 0.002457
±
1.542
0 0 0
0 196952 3048
0 0 0
θ
: #Hits MVD vs.
-
K
2015-01-25 17:22:21
θ
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
#Hits
0
10
20
30
40
50
60
70
80
Entries 200000
Mean x 0.0004736
±
0.354
Mean y 0.02423
±
12.16
RMS x 0.0003349
±
0.2102
RMS y 0.01713
±
10.75
0 0 0
0 196952 3048
0 0 0
0
500
1000
1500
2000
2500
3000
3500
Entries 200000
Mean x 0.0004736
±
0.354
Mean y 0.02423
±
12.16
RMS x 0.0003349
±
0.2102
RMS y 0.01713
±
10.75
0 0 0
0 196952 3048
0 0 0
θ
: #Hits STT vs.
-
K

View Slide

89
D Meson Production Cross Section
6.0 6.5 7.0 7.5 8.0
p
lab
(GeV/c)
6.0 6.5 7.0 7.5 8.0
p
lab
(GeV/c)
10-3
10-2
10-1
100
σ
tot
(µb)
pp -> D+
D-
FIG. 3: Total reaction cross sections for ¯
pp → D ¯
D as a func-
tion of plab
, based on baryon-exchange (shaded band) and
the quark model (grid). Results obtained in Born approx-
imation are indicated by the dotted (baryon-exchange) and
Baryon Exchange Model
Quark Model
p
lab
(GeV/c)
6.0 6.5 7.0 7.5 8.0
p
lab
(GeV/c)
10-3
10-2
10-1
100
σ
tot
(µb)
pp -> D+
D-
3: Total reaction cross sections for ¯
pp → D ¯
D as a func-
of plab
uark model (grid). Results obtained in Born approx-
on are indicated by the dotted (baryon-exchange) and
dotted (quark model) lines, respectively.
lab
6.0 6.5 7.0 7.5 8.0
p
lab
(GeV/c)
10-3
10-2
10-1
100
σ
tot
(µb)
pp -> D+
D-
3: Total reaction cross sections for ¯
pp → D ¯
D as a func-
of plab
uark model (grid). Results obtained in Born approx-
on are indicated by the dotted (baryon-exchange) and
dotted (quark model) lines, respectively.
J. Haidenbauer, G. Krein; Production of charmed pseudoscalar mesons
in antiproton-proton annihilation; arXiv:1404.4174 [hep-ph] (04-2014)
from Foley et al. [43], Berglund et al. [47], Russ et al. [48],
Λ,Σ,
Σ(1385)
K K
N N
✉ ✉ ↔
Λc
,Σc
,
Σc
(2520)
D D
N N
✉ ✉
FIG. 2: Transition potential for ¯
NN → D ¯
D (right) and
¯
NN → ¯
KK (left), respectively.
D ¯
D, ¯
NN = V D ¯
D, ¯
NN
+ V D ¯
D, ¯
NN G ¯
NN T ¯
NN, ¯
NN , (5)
he ¯
NN potential described in Sect. II.
Eq. (5) implies that the ¯
NN → D ¯
D transition
s effectively evaluated in a DWBA.
with inclusion of ISI effects are presented as
ig. 3 because we consider several variants of
otential as discussed in the previous section.
s that the results change drastically once the
ded in the calculation. The cross sections for
trongly reduced while at the same time those
are enhanced. Indeed now both D ¯
D channels
d at a comparable rate. In fact, the predicted
n for D+D− appears to be even somewhat
the one for D0 ¯
D0.
the reduction in the D0 ¯
D0 case is in line
arable effects observed in the previous stud-
annihilation processes [23, 25–27], as men-
ve, the enhancement seen for D+D− may be
6.0 6.5 7.0 7.
p
lab
(GeV/c)
10-3
10-2
10-1
100
σ
tot
(µb)
pp
tion of plab
, based on baryon-exchange (s
the quark model (grid). Results obtained
imation are indicated by the dotted (bary
dash-dotted (quark model) lines, respective
Q Q
q q q q
FIG. 4: Microscopic quark-model mechanism for the
tion potential: annihilation of two pairs of light quark
u¯
u, d ¯
d, and creation of a pair of heavier quarks, Q ¯
Q =
(¯
cc) is created – see Fig. 4. We base our study
model of Kohno and Weise (KW) [28] for the ¯
pp
reaction; we replace parameters corresponding
s−quark and K−meson of that model by those
c−quark and D−meson. The quark-model ¯
NN
transition potential V ¯
NN→D ¯
D
Q
(t) can be written
V ¯
NN→D ¯
D
Q
(t) = χ†
¯
N
[h1
(t) σ · p + h2
(t) σ · p] χN
cutoff mass below.
us now focus on the effects of the initial state inter-
Those effects are included by solving the formal
d-channel equations
T ¯
NN, ¯
NN = V ¯
NN, ¯
NN
+ V ¯
NN, ¯
NN G ¯
NN T ¯
NN, ¯
NN , (4)
T D ¯
D, ¯
NN = V D ¯
D, ¯
NN
+ V D ¯
D, ¯
NN G ¯
NN T ¯
NN, ¯
NN , (5)
ng the ¯
NN potential described in Sect. II.
rse, Eq. (5) implies that the ¯
NN → D ¯
D transition
ude is effectively evaluated in a DWBA.
ults with inclusion of ISI effects are presented as
in Fig. 3 because we consider several variants of
N potential as discussed in the previous section.
bvious that the results change drastically once the
ncluded in the calculation. The cross sections for
are strongly reduced while at the same time those
+D− are enhanced. Indeed now both D ¯
D channels
oduced at a comparable rate. In fact, the predicted
section for D+D− appears to be even somewhat
than the one for D0 ¯
D0.
ereas the reduction in the D0 ¯
D0 case is in line
omparable effects observed in the previous stud-
¯
NN annihilation processes [23, 25–27], as men-
above, the enhancement seen for D+D− may be
6.0 6.5 7.0 7.5
p
lab
(GeV/c)
10-3
10-2
6.0 6.5 7.0 7.5
p
lab
(GeV/c)
10-3
10-2
10-1
100
σ
tot
(µb)
p
pp
tion of plab
, based on baryon-exchange (sh
the quark model (grid). Results obtained
imation are indicated by the dotted (baryo
dash-dotted (quark model) lines, respectivel
82

View Slide

GPU-based Online Tracking Algorithms & Application to D±→K∓π±π±

GPU-based Online Tracking Algorithms & Application to D±→K∓π±π±

AndiH

More Decks by AndiH

Other Decks in Science

Featured

Transcript