Tracking in LHCb's 2020 HLTConnecting the Dots 2015
Tim Head, on behalf of LHCb
École Polytechnique Fédérale de Lausanne
9 February 2015
The LHCb Detector
250mrad
100mrad
• LHCb is a single-arm (2 < η < 5) spectrometer at the LHC
I Precision beauty and charm physics: CP violation, rare decays, heavy flavourproduction
• σbb and σcc are extremely large at the LHC
I 30 kHz bb̄ and 600 kHz cc̄ in LHCb acceptance!
Phys.Lett.B694(2010)209-216
Eur.Phys.J.C71
(2011)1645
Tim Head (EPFL) 9 February 2015 2
Heavy Flavour Signatures
Beauty hadrons
• B+ mass 5.28GeV, daughter pTO (1 GeV)
• lifetime ≈ 1.6ps ⇒ flight distance
≈ 1 cm
• common signature: detached µµ
Charmed hadrons
• D0 mass 1.86GeV, sizeable daughter pT
• lifetime ≈ 0.4ps ⇒ flight distance
≈ 4mm
• can be produced in B decays
Tim Head (EPFL) 9 February 2015 3
Actual Signatures
SV
h
h
IP
Topo N-body
Inclusive dimuon
SVSV
p p
p pIP
IPp p
PV
hh
D∗ → πD(→hh)
p pπ±
Inclusive φµ+
µ−
B → hh
p p
h
h
K−K+
• Combining inclusive and
exclusive selections
• Main trigger is a Bonsai BDT
• What you really need is offline
like event reconstruction!
• Extremely powerful and
flexible software environment
Tim Head (EPFL) 9 February 2015 4
LHCb at 40MHz
• In Run 2 LHCb will collect ≈ 8 fb−1 ⇒increase instantaneous luminosity
• At increased luminosity signals less well
separated in L0 ⇒ we need to read out
every event!
• Upgrade readout to 40MHz, full detector
readout of all visible pp interactions
• Replace hardware L0 by software Low Level
Trigger (LLT)
I Acts as temporary ``handbrake'' duringcommissioning, 1− 40MHz scaleableoutput rate
Tim Head (EPFL) 9 February 2015 5
The Game has ChangedIn the upgrade area there are no ``boring'' events, it is about classifying signal events!
80GB/s of reconstructible D hadrons, 27 GB/s of reconstructible B hadrons. Compare to
10GB/s allowed to tape
Details: LHCb-PUB-2014-027
Tim Head (EPFL) 9 February 2015 6
The Game has ChangedIn the upgrade area there are no ``boring'' events, it is about classifying signal events!
80GB/s of reconstructible D hadrons, 27 GB/s of reconstructible B hadrons. Compare to
10GB/s allowed to tape
Details: LHCb-PUB-2014-027
83
Triggersin the future
Triggerstoday
Gligorov 2014Tim Head (EPFL) 9 February 2015 6
Real Time Analysis
Tim Head (EPFL) 9 February 2015 7
The LHCb Experiment in 2018
250mrad
100mrad
UT
SciFi
CER
N-L
HC
C-2
013-
021
/LH
CB-
TDR
-013
05/1
2/20
13
CER
N-L
HC
C-2
013-
022
/LH
CB-
TDR
-014
05/1
2/20
13
TDRUPGRADE
CERN/LHCC 2013-022
LHCb TDR 14
28 November 2013
LHCbLHCbParticle IdentificationParticle Identification
Technical Design Report
TDRUPGRADE
CERN/LHCC 2014-001
LHCb TDR 15
21 February 2014
LHCbTracker
Technical Design Report
Upstream Tracker
SciFi Tracker
LHCB-TDR-013
LHCB-TDR-014
LHCB-TDR-015
Tim Head (EPFL) 9 February 2015 8
The LHCb Experiment in 2018
250mrad
100mrad
UT
SciFi
CER
N-L
HC
C-2
013-
021
/LH
CB-
TDR
-013
05/1
2/20
13
CER
N-L
HC
C-2
013-
022
/LH
CB-
TDR
-014
05/1
2/20
13
TDRUPGRADE
CERN/LHCC 2013-022
LHCb TDR 14
28 November 2013
LHCbLHCbParticle IdentificationParticle Identification
Technical Design Report
TDRUPGRADE
CERN/LHCC 2014-001
LHCb TDR 15
21 February 2014
LHCbTracker
Technical Design Report
Upstream Tracker
SciFi Tracker
CER
N-L
HC
C-2
014-
016
/LH
CB-
TD
R-0
1622
/05/
2014
TDRUPGRADE
CERN/LHCC 2014-016
LHCb TDR 16
21 May 2014
LHCbTrigger and Online
Technical Design Report
LHCB-TDR-013
LHCB-TDR-014
LHCB-TDR-015
LHCB-TDR-016
Tim Head (EPFL) 9 February 2015 8
Vertex Locator (VELO)
• Hybrid pixel detectors, two moveable halves, active edge at 5.1mm from beam
• Basic building blocks are 14× 14mm2 pixel chips, three chips in a row share silicon
sensor
• Micro-channel CO2 cooling
UpgradeCurrent
Tim Head (EPFL) 9 February 2015 9
Upstream Tracker
• Single-sided silicon strip detector
• Four layers (x, u, v, x) (5° stereo angles)
• Finer segmentation around beam-pipe
• 250µm thin sensors
• New read-out chip (SALT)
• Bi-phase CO2 cooling
Tim Head (EPFL) 9 February 2015 10
Scintillating Fibre Tracker
• Scintillating fibres
I 250µm diameter, 2.5m long
• Three stations with four (x, u, v, x) layers each
• Read out by Silicon Photomultipliers
I inside light-tight read-out boxI cooled to −40 °C
• New ASIC for read-out (PACIFIC)
250µm
Tim Head (EPFL) 9 February 2015 11
Tracking at 30MHz
VELO
Upstream Tracker
SciFi
Tim Head (EPFL) 9 February 2015 12
Tracking at 30MHz
• Reconstruct all tracks
• Build doublets, extrapolate
to next module
• Extremely efficient and ghost
free
Velo
Tim Head (EPFL) 9 February 2015 12
Tracking at 30MHz
from E. Bowen
Tim Head (EPFL) 9 February 2015 12
Tracking at 30MHz
• Extrapolate tracks with
pT > 200MeV
• Measure track's curvature
• σp/p ≈ 15%
VeloUT
Tim Head (EPFL) 9 February 2015 12
Tracking at 30MHz
Tim Head (EPFL) 9 February 2015 12
Tracking at 30MHz
• Extrapolate tracks with
pT > 500MeV
• Use UT charge estimate to
cut search windows in half
• σp/p ≈ 0.5%
LongLong
Tim Head (EPFL) 9 February 2015 12
Full Track Reconstruction on all Events
Velo tracking
Offline Tracking
Forward trackingpT > 70 MeV, δp/p ~ 0.5%
PV finding
Velo tracking
Online Tracking
Forward trackingpT > 500 MeV, δp/p ~ 0.5%
PV finding
Simplified Kalman fit
Particle Identification
Velo-UT trackingpT > 200 MeV, δp/p ~ 15%
Rate reducing cutsOutput < 1 MHz
Muon Identification
Full Kalman fit
Particle Identification
• Maximum flexibility and
robustness
• Details: LHCb-PUB-2014-028
• LHCb will be the first hadron
collider experiment to operate a
software only trigger at full event
rate!
Offline-quality tracking at 30MHz in software is possible!
Tim Head (EPFL) 9 February 2015 13
Efficiency
• Compared to ``offline'' the HLT tracking
sequence is 98.7% efficient
• In addition tracks with pT < 0.5GeV/care available with lower momentum
resolution
Efficiency [%]
HLT relative
long, from B 72.8 80.3
long, pT > 0.5GeV/c 87.4 97.2
long, from B, pT > 0.5GeV/c 92.5 98.7
]-1c[GeVTp
0 5 10
Eff
icie
ncy
0.80
0.85
0.90
0.95
1.00
1.05LHCb simulation
Trigger
Offline
Ratio
Tim Head (EPFL) 9 February 2015 14
Timing
• At nominal luminosity reconstruction
uses less than half the budget (13ms)
• CPU time does not ``explode'' at higher
luminosity
Algorithm CPU time [ms]
VELO 2.0
VELO-UT 1.3
Forward 1.9
PV finding 0.38
Total 5.4 3.8 7.6 11.4ν
0
1
2
3
4
5
6
7
CPU
tim
e [m
s]
LHCb simulation
VELO trackingVELO-UT trackingForward trackingPV finding
(ν =average number of interactions)
Tim Head (EPFL) 9 February 2015 15
Conclusion
• The LHCb trigger has been very successful in 2011 and 2012
I using BDTs as the main trigger
• 2020 will see a truly upgraded trigger
• Tracking at 30MHz in software is possible
• Allows very diverse, efficient triggers that minimally bias the physics observables
I lifetime unbiased hadronic triggers
• Run 2 will be our ``test beam'', testing many techniques which will be needed for
2020:
I run-by-run calibration and alignmentI Turbo stream: analysis without offline reconstruction
Tim Head (EPFL) 9 February 2015 16
One more thing ...
HLT1 Tracking Time
• Making the whole tracking sequence
faster is hard work
• Some competition is good as nobody
wants to be the slowest
• A dedicated group of people, working
together is needed
Jul 20
13
Aug 20
13
Sep 20
13
Oct 20
13
Nov 20
13
Dec 20
13
Jan 20
14
Feb 20
14
Mar 20
14
Apr 20
140
2
4
6
8
10
12
14
Trig
ger t
rack
ing
time
[ms]
LHCb simulation
Spillover was turned on
Total with GEC
VeloVeloUTForward
Tim Head (EPFL) 9 February 2015 18
SciFi Track Reconstruction
• Forward track reconstruction algorithm efficiency and ghostrate
• Taken from LHCB-TDR-015
Tim Head (EPFL) 9 February 2015 19
Top Related