B. Storaci, CERN Seminar

First full calibration and
reconstruction of a HEP detector
in real time
Barbara Storaci on behalf of the LHCb Collabora3on Detector Seminar, October 9th 2015 Layout
•  The LHCb experiment •  The challenge: doing a full calibra3on and reconstruc3on in real-­‐3me –  New resources –  Improved reconstruc3on chain –  Calibra3on and alignment on the online farms •  How it converts into physics: –  The Turbo stream and the early measurements (see also the talk of Patrick Spradlin, Wednesday 7th October) Barbara Storaci, University of Zurich 2 The LHCb experiment
•  LHCb is the dedicated heavy flavor physics experiment at LHC •  Single arm forward spectrometer, covering a pseudorapidity range unique among the LHC detectors •  Its primary goal is to look for indirect evidence of new physics in CP-­‐viola3on and rare decays of beauty and charms hadrons •  This requires: 1. 
2. 
3. 
Excellent tracking (momentum, impact parameters and primary vertex resolu3on) Excellent decay 3me resolu3on Excellent par3cle iden3fica3on Barbara Storaci, University of Zurich 3 The tracking systems
Barbara Storaci, University of Zurich 4 Vertex Locator (VELO)
• 
• 
• 
• 
Vertex Locator (VELO): 42 silicon micro-­‐strip sta3ons with r-­‐Phi sensors Surrounding interac3on point Two retractable halves, 8 mm from beam when closed •  Closed for each fill Possibility to inject gas (like He, Ne, Ar) for special data taking Performance of the LHCb Vertex Locator JINST 9 (2014) 09007 LHCb Detector Performance Int.J.Mod.Phys. A30 (2015) 1530022 Vertex Locator (VELO)
• 
• 
• 
• 
Vertex Locator (VELO): 42 silicon micro-­‐strip sta3ons with r-­‐Phi sensors Surrounding interac3on point Two retractable halves, 8 mm from beam when closed •  Closed for each fill Possibility to inject gas (like He, Ne, Ar) for special data taking Performance of the LHCb Vertex Locator JINST 9 (2014) 09007 LHCb Detector Performance Int.J.Mod.Phys. A30 (2015) 1530022 VELO performance in Run I
PV resoluGon of ~12μm in x/y ~71μm in z Decay Gme resoluGon of 45 fs for a 4-­‐tracks vertex JINST 9 (2014) 09007 Int.J.Mod.Phys. A30 (2015) 1530022 Tracker Turicensis (TT)
LHCb Detector Performance Int.J.Mod.Phys. A30 (2015) 1530022 Tracker Turicensis (TT): •  Four planes (0o,+5o,-­‐5o,0o) of silicon micro-­‐
strip sensor •  Total silicon area of 8m2 •  The detector operate at 0oC. •  Upstream of the magnet •  Already sensiGve to the magneGc field Inner and Outer Trackers (IT and OT)
Outer trackers (OT) •  Downstream of the magnet •  Three sta3ons each with four planes of straw tubes •  Gas Mixture Ar/CO2/O2 (70/28.5/1.5) Inner Tracker (IT): •  Downstream of the magnet •  Three sta3ons each with four planes of silicon micro-­‐strip sensors around the beam pipe •  Total silicon area of 4.2m2 •  The detector operate at 0oC. Performance of the LHCb Outer Tracker JINST 9 (2014) P01002 LHCb Detector Performance Int.J.Mod.Phys. A30 (2015) 1530022 9 Tracking performance
•  Stable tracking efficiency in RunI •  Above 96% in all the momentum range for track traversing the full detector Int.J.Mod.Phys. A30 (2015) 1530022 The PID systems
Barbara Storaci, University of Zurich 11 The PID systems
RICH1 RICH2 spherical mirrors Hybrid Photon Detectors Ring Imaging CHerenkov detectors (RICH1 and RICH2) •  RICH1: •  RICH2: •  Upstream of the magnet •  Downstream of the magnet •  C4F10 radiator •  CF10 radiator •  2<p<40 GeV/c •  15<p<100 GeV/c •  25-­‐300 mrad •  15-­‐120 mrad Performance of the LHCb RICH detector Eur.Phys.J. C73 (2013) 2431 LHCb Detector Performance Int.J.Mod.Phys. A30 (2015) 1530022 The trigger and PID systems
Barbara Storaci, University of Zurich 13 The trigger and PID systems
Muon mulG-­‐wire proporGonal chambers Performance of the Muon IdenGficaGon system JINST 8 (2013) P10020 ECAL Electromagne3c and hadronic calorimenters (ECAL and HCAL) •  Scin3llator planes + absorber material planes •  Provide the Level 0 signature for events with hadrons above a certain ET •  Offline PID for photons, electrons and hadrons Muon system •  5 sta3ons, each equipped with 276 mul3-­‐wire propor3onal chambers •  Inner part of the first sta3on equipped with 12 GEM detectors •  Level 0 trigger selec3on of tracks with high pT •  Offline PID of muons LHCb Detector Performance 14 Int.J.Mod.Phys. A30 (2015) 1530022 The PID performances
Kaon idenGficaGon efficiency ~95% with a pion mis-­‐idenficaGon frac3on of ~10% over the full 2-­‐100GeV/c momentum range (3ght selec3on) Capability to dis3nguish par3cles (especially π-­‐K) in a the momentum range 2-­‐100 GeV/c) Electron idenGficaGon efficiency >91% with a hadron mis-­‐idenfica3on frac3on of few % p>10 GeV/c (intermediate selec3on) Int.J.Mod.Phys. A30 (2015) 1530022 Eur.Phys.J. C73 (2013) 2431 15 Trigger in RunI
•  Level 0: 1MHz rate –  Implemented in hardware –  High pT and ET signature in muon and calorimeter systems •  Higher Level Trigger (HLT): 5kHz rate –  Flexible sojware triggers with two stages (HLT1 and HLT2) –  Track reconstruc3on and PV finding performed •  Simplified reconstruc3on w.r.t. offline •  Preliminary alignment and calibra3on •  RICH PID info marginally used (not-­‐
fully calibrated system) •  Processing of the data at the end of the year with the latest constants •  Only 20% deferred to disk JINST 8 (2013) P04022 16 Idea
Novel concept in HEP: Being able to do physics directly on the HLT output Advantages: •  No need of offline data processing (data available ~ immediately ajer HLT2 has processed them) •  Raw event size can be much smaller à possibility to reduce the pre-­‐scaling of high branching ra3o channels (like for charm physics) Do much more physics with the given resources! Barbara Storaci, University of Zurich 17 Trigger in RunII
•  Events from lower trigger levels can be buffered on disk while performing real-­‐Gme alignment and calibraGon •  Last trigger level uses the same reconstrucGon as offline •  Same alignment and calibraGon constants used by the trigger and the offline reconstruc3on •  Some analysis performed directly on the trigger output Barbara Storaci, University of Zurich 18 New resources
•  HLT farm nearly doubled from Run I: –  ~27k physical cores –  Farm nodes added for RunII are about 2x more powerful than previous nodes •  Event can be buffered ajer HLT1: 5PB disk space SGll a difficult challenge: •  Managing to do the full reconstruc3on in few hundreds ms achieving offline performance (same or bener than in RunI) •  Managing to do the full alignment and calibra3on of the detectors in real-­‐3me •  Being able to select efficiently many signals already at trigger level Barbara Storaci, University of Zurich 19 Reconstruction chain optimization
Barbara Storaci, University of Zurich 20 Optimization of the reconstruction chain
•  In order to maintain full reconstruc3on efficiency, while keeping within a strict 3ming budget many improvements were needed: 1.  Op3miza3on of the code (e.g. vectoriza3on) 2.  Changes to the reconstruc3on chain and op3miza3on of the panern algorithm Optimization of the reconstruction chain
1.  OpGmizaGon of the code (e.g. vectorizaGon): Examples •  Iden3fied hot spots by profiling •  VectorisaGon (track fit, magneGc field) •  Caching (material descrip3on) •  Fast approxima3ons •  Algorithms tuning and re-­‐implementa3on => Possible to gain order of ~30% in several algorithms Optimization of the reconstruction chain
2.  Changes to the reconstrucGon chain and opGmizaGon of the pagern algorithm: •  Velo tracks can be extended to the TT detector Advantages: •  The TT detector is already in the magne3c field: possible to es3mate q/p (resolu3on ~15%) •  Momentum es3mate used to pre-­‐
select tracks at an early stage •  Charge es3mate allows greatly reduced search windows downstream of the magnet Example Optimization of the reconstruction chain
LHCb simula3on LHCb simula3on VELO-­‐TT performance: •  More than 97% efficient for tracks with pT>200 MeV that have hits in at least three layers •  ~5% beger performance than previous code (never used since it was too slow) New reconstrucGon chain: •  ~3 Gmes faster •  Bener or equivalent performance as in Run I •  Tracks with a pT>500MeV already available at HLT1 level (in Run I pT>1.3 GeV) without any IP requirements needed. •  Gains >50% signal efficiency for charm physics •  Enable lifeGme unbiased triggers for hadronic final states (world first!) The new chain at work
100
•  IP resoluGon comparable with 2012 data •  Stable between different fills •  Compa3ble results with 50 and 25 ns data 90
2012 Data
80
2015 25 ns Data
X
σ(IP ) [µm]
70
60
50
40
30
LHCb VELO Preliminary
2012 Data: σ = 11.6 + 23.4/p
20
T
10
0
0
2015 25 ns Data: σ = 11.8 + 23.7/p
T
0.5
1
1.5
2
2.5
3
1/p [c/GeV]
1.05
Resolution Δx [mm]
ε
T
LHCb preliminary
1
0.95
0.9
0.85
Data EM 2015
0.8
0.04
LHCb Simulation
0.03
Online Run II
0.02
Offline Run I
0.01
Data 2012
0.75
50
100
150
200
p [GeV/c ]
Tracking efficiency comparable with 2012 data 0
20
40
60
80
Multiplicity of reconstructed tracks
IdenGcal resoluGon to Run I for the PV reconstrucGon with 70% smaller fracGon of fake primary verGces SGll a difficult challenge: •  Managing to do the full reconstruc3on in … achieving offline performance (same or bener than in RunI): DONE! •  Managing to do the full alignment and calibra3on of the detectors in real-­‐3me •  Being able to select efficiently many signals already at trigger level Barbara Storaci, University of Zurich 26 Real-time alignment and calibration
Barbara Storaci, University of Zurich 27 Importance of the alignment (I)
•  Physics performance relies on the spa3al alignment of the detector and the accurate calibra3on of its subcomponents: 1. 
Accurate alignment of the VELO essen3al for primary ver3ces discrimina3on, excellent impact parameter (IP) and proper 3me resolu3on IPx resoluGon vs 1/pT First alignment σIP (high pT) = 14.0 μm Int.J.Mod.Phys. A30 (2015) 1530022 Latest alignment σIP (high pT) = 11.6 μm 28 Importance the alignment (II)
•  Physics performance relies on the spa3al alignment of the detector and the accurate calibra3on of its subcomponents: 2.  Bener alignment of the tracking system improves the mass resolu3on Invariant mass distribuGon for ϒèμ+μ-
First alignment Bener alignment σϒ = 92 MeV/c2 σϒ = 49 MeV/c2 LHCb Preliminary Barbara Storaci, University of Zurich LHCb Preliminary 29 Importance of the calibration
•  Complete calibra3on of the RICH detectors needed for exclusive selec3on using hadron par3cle iden3fica3on criteria JHEP 1210 (2012) 037 Barbara Storaci, University of Zurich 30 Real-time Alignment and Calibration(I)
•  General Strategy –  Automa3c evalua3on at regular intervals, e.g. per fill or per run depending on the task –  Dedicated data sample to perform alignment or calibra3on collected with specific trigger selec3on line for each task –  Compute the new alignment or calibra3on constants in few minutes –  Update the constants only if needed –  The same new alignment and calibraGon constants will be used both by the trigger and the offline reconstrucGon • 
• 
• 
• 
Advantages: Have the same performance online and offline More effec3ve trigger selec3on Stability of the alignment quality, hence physics performance Some analysis performed directly on the trigger output 31 Real-time Alignment and Calibration(II)
•  Technicali3es: –  Constants expected to change: updated in real-­‐3me (each fill, run, …) –  Constants expected to be stable: monitoring •  Took advantage of the farm compu3ng power CalibraGon Farm tasks: •  RICH refrac3ve index and HPD image calibra3on •  Calorimeter calibra3on •  OT-­‐t0 calibra3on Online Farm tasks: • 
• 
• 
• 
VELO and tracker aligment Muon alignment RICH mirror alignment Calorimeter π0 calibra3on Barbara Storaci, University of Zurich 32 Real-time Alignment and Calibration(II)
CalibraGon Farm Online Farm •  Tasks are run on a single node •  Evalua3on of the parameters by fitng online-­‐monitoring histogram Tasks: •  RICH refrac3ve index and HPD image calibra3on •  Calorimeter calibra3on •  OT-­‐t0 calibra3on Barbara Storaci, University of Zurich 33 Real-time Alignment and Calibration(II)
CalibraGon Farm • 
• 
• 
• 
• 
Online Farm Tasks are run on a single node •  Parallel processing on ~1700 HLT farm Evalua3on of the parameters by nodes fitng monitoring histogram •  Evalua3on of the parameters by itera3ve process Tasks: •  Analyser (mul3ple nodes): RICH refrac3ve index and HPD perform reconstruc3on image calibra3on •  Combina3on of output, fits/
Calorimeter calibra3on minimiza3on -­‐> extract constants OT-­‐t0 calibra3on Tasks: •  VELO and tracker aligment •  Muon alignment •  π0-­‐calibra3on •  RICH mirror alignment Barbara Storaci, University of Zurich 34 Real-time Alignment and Calibration(II)
•  Technicali3es: –  Constants expected to change: updated in real-­‐3me (each fill, run, …) –  Constants expected to be stable: monitoring •  Took advantage of the farm compu3ng power CalibraGon Farm tasks: •  RICH refrac3ve index and HPD image calibra3on •  Calorimeter calibra3on •  OT-­‐t0 calibra3on Online Farm tasks: • 
• 
• 
• 
VELO and tracker aligment Muon alignment RICH mirror alignment Calorimeter π0 calibra3on Barbara Storaci, University of Zurich 35 VELO, Tracker and Muon alignment:
logic
²  Uses an itera3ve procedure to minimise the residuals of a Kalman fit to a sample of reconstructed tracks ²  Mul3ple scanering and magne3c field are taken into account and both par3cle masses and informa3on from ver3ces are used as global constraints Iterate un3l χ2 difference is below threshold Barbara Storaci, University of Zurich 36 VELO, Tracker and Muon alignment:
at work
•  Automa3c alignment procedure runs at the start of each fill è update of the constants only if necessary –  VELO: is opened and closed every few fills. •  Expected to update constants every few fills –  Tracking systems (TT, IT, OT) •  Expected to update constants every few weeks –  Muon system Variation [µm]
LHCb VELO
x-translation
y-translation
Preliminary
200
150
LHCb Tracker
100
Preliminary
IT1 ASide
IT1 Bottom
IT1 Top
IT1 CSide
50
0
−50
−100
−150
05/07/2015 - 01/10/2015
−200
05/07/2015 - 01/10/2015
4444
4440
4434
443228
44
442263
44
4420
4410
4402
438861
43
4360
434491
43
433273
43
4322
4266
4257
4256
4249
424463
42
4231
4225
4220
4214
421121
42
421008
42
420071
42
401098
40
4006
3996
399828
39
398863
39
3981
3976
397741
39
3965
3962
4444
4440
4437
4435
4434
4432
4428
4426
4423
4420
4418
4410
4402
4398
4393
4386
4381
4349
4341
4337
4322
4257
4256
4249
4246
4243
4231
4225
4220
4214
4212
4211
4210
4208
4205
4019
4008
3996
3992
3988
3981
3976
3974
3971
3965
3962
10
8
6
4
2
0
−2
−4
−6
−8
−10
ΔX Variation [µm]
•  Only as monitoring (expected only ajer hardware interven3ons to the system) Fill number
Fill number
VELO, Tracker and Muon alignment:
RunI vs Run II
• 
• 
• 
• 
• 
• 
• 
• 
Run I Run offline few 3mes per year (and updated online) Time needed to run offline ~1h Data taken with a previous version of the alignment -­‐> not best condi3ons used in HLT New alignment updated offline for the re-­‐processing of the data Run II Run every fill and constants automaGcally updated Time needed ~7 minutes HLT always with the most updated constants Offline synchronized with online parameters Barbara Storaci, University of Zurich 38 Real-time Alignment and Calibration(II)
•  Technicali3es: –  Constants expected to change: updated in real-­‐3me (each fill, run, …) –  Constants expected to be stable: monitoring •  Took advantage of the farm compu3ng power CalibraGon Farm tasks: •  RICH refrac3ve index and HPD image calibra3on •  Calorimeter calibra3on •  OT-­‐t0 calibra3on Online Farm tasks: • 
• 
• 
• 
VELO and tracker aligment Muon alignment RICH mirror alignment Calorimeter π0 calibra3on Barbara Storaci, University of Zurich 39 RICH Mirror alignment: logic
•  Cherenkov photons focused on photon-­‐detector plane by spherical and flat mirrors •  An imaginary track reflected through the RICH mirrors should be in the center of the Cherenkov ring •  The distribu3on of the Δθ against ϕ results in a sinusoidal distribu3on in case of misalignment of the mirrors •  Fit the distribu3on to calculate alignment constants 40 RICH mirror alignment: at work
•  Mirror pairs to align: Before alignment –  RICH1: 16 –  RICH2: 94 •  1090 alignment constants •  Re-­‐opGmizaGon of the code and HLT-­‐
selec3on of the high momentum tracks to be run in the online environment •  Same framework as the tracking alignment: Amer alignment –  Analysers: photon reconstruc3on done in parallel –  Iterator: fit of the Δθ distribu3on on a single node •  Used as monitoring (if enough sta3s3cs available poten3ally each fill) 41 RICH mirror alignment: RunI vs Run II
• 
• 
• 
• 
Run I Run offline few 3mes per year Time needed to run offline ~several days on the GRID Minimal usage of RICH informa3on at HLT level (not fully calibrated system) New alignment updated offline for the re-­‐stripping of the data Run II • 
• 
• 
• 
• 
Re-­‐op3mized code Run as soon as enough staGsGcs is accumulated (typically 1-­‐2 fills) Time needed ~30 minutes RICH informa3on can be used in HLT: fully calibrated system Used as monitoring Barbara Storaci, University of Zurich 42 Real-time Alignment and Calibration(II)
•  Technicali3es: –  Constants expected to change: updated in real-­‐3me (each fill, run, …) –  Constants expected to be stable: monitoring •  Took advantage of the farm compu3ng power CalibraGon Farm tasks: •  RICH refrac3ve index and HPD image calibra3on •  Calorimeter calibra3on •  OT-­‐t0 calibra3on Online Farm tasks: • 
• 
• 
• 
VELO and tracker aligment Muon alignment RICH mirror alignment Calorimeter π0 calibra3on Barbara Storaci, University of Zurich 43 RICH calibration: logic
•  RICH automa3c calibra3on consists of calibra3ng: 1.  RICH refrac3ve index: •  Depends on the gas mixture, temperature and pressure •  Determined by fitng the difference between reconstructed and expected Cherenkov angle –  Hybrid photon detector (HPD) images •  Affected fields distor3ons Before alignment by the magne3c and electric Amer alignment •  Anode images cleaned and Sobel filter is used to detect the edges Rich2Gas Rec-Exp Cktheta | All photons
3
×10
RICH/RiCKResLongTight/Rich2Gas/ckResAll
Entries
100
Mean
Std Dev
1.451864e+07
−2.706e−05
0.001322
90
80
70
60
LHCb Preliminary LHCb Preliminary 50
−0.002
Barbara Storaci, University of Zurich −0.001
0
0.001
0.002
44 RICH calibration: logic
•  RICH automa3c calibra3on consists of calibra3ng: 2.  Hybrid photon detector (HPD) images •  Affected by the magne3c and electric fields distor3ons •  Anode images cleaned and Sobel filter is used to detect the edges –  RICH refrac3ve index: •  Depends on the gas mixture, temperature and pressure •  Determined by fitng the difference between reconstructed and expected Cherenkov angle Barbara Storaci, University of Zurich 45 RICH calibration: at work
Fully automa3c procedure, evaluated run by run •  In case of a too short run, the same constants of the previous run are used •  Calibra3on parameters are monitored by PVSS trend plots •  Maintained stable condi3ons during 2015 data taking RICH 1 RICH 2 σΔθ=1.641 mrad σΔθ=0.67 mrad LHCb Preliminary LHCb Preliminary Real-time Alignment and Calibration(II)
•  Technicali3es: –  Constants expected to change: updated in real-­‐3me (each fill, run, …) –  Constants expected to be stable: monitoring •  Took advantage of the farm compu3ng power CalibraGon Farm tasks: •  RICH refrac3ve index and HPD image calibra3on •  Calorimeter calibra3on •  OT-­‐t0 calibra3on Online Farm tasks: • 
• 
• 
• 
VELO and tracker aligment Muon alignment RICH mirror alignment Calorimeter π0 calibra3on Barbara Storaci, University of Zurich 47 Calorimeter calibration: logic
•  Important to have stable L0 rates •  It allows to account for ageing effects •  The calibra3on adjust directly the HV setngs •  Absolute calibra3ons: –  π0 calibra3on: •  Selected photons (3x3 clusters) and fix seed (central) cell –  Compute di-­‐photon invariant mass –  Find the coefficient which would move the measurement closer to the nominal π0 mass –  Cs source scan: •  Scan of the response with the source. Done each TS •  Rela3ve calibra3ons: –  Raw occupancy method: •  Comparison of the performance of each cell with a reference –  LED monitoring system: •  To detect ageing of the PMTs Barbara Storaci, University of Zurich 48 Calorimeter calibration: at work
• 
The LED correc3ons for ECAL and HCAL are automa3cally performed/updated every fill –  ~15 minutes to have the correc3ons available • 
• 
Commissioning of the occupancy method ongoing: wai3ng for stable beam condi3ons (not available in RunI) π0 calibra3on done every months: now it takes only few hours to produce the results π0-­‐calibraGon LHCb Preliminary LED average over 456 cells in the very central part of ECAL 2012 LHCb Preliminary 2015 LHCb Preliminary 2012 – no correc3ons 2015: green line – π0 based calibra3on; blue lines – LED based correc3ons Real-time Alignment and Calibration(II)
•  Technicali3es: –  Constants expected to change: updated in real-­‐3me (each fill, run, …) –  Constants expected to be stable: monitoring •  Took advantage of the farm compu3ng power CalibraGon Farm tasks: •  RICH refrac3ve index and HPD image calibra3on •  Calorimeter calibra3on •  OT-­‐t0 calibra3on Online Farm tasks: • 
• 
• 
• 
VELO and tracker aligment Muon alignment RICH mirror alignment Calorimeter π0 calibra3on Barbara Storaci, University of Zurich 50 OT-t0 calibration: logic
JINST 9 (2014) P01002 •  Global 3me offset (t0) related to the synchroniza3on between the OT 3me and the collision 3me -­‐> a shij of 0.5 ns leads to tracking inefficiency of ~0.25% •  Calculate t0 for the en3re OT (real-­‐3me update) •  The offset is evaluated by using the reconstructed tracks, and comparing the drij 3me with the expected drij 3me (evaluated using the TR-­‐rela3on) LHCb Good tracks/ (0.4 ns)
t = tmeas
t(r)
×106
18 LHCb OT
16
14
12
10
8
6
4
2
0
-20
-10
0
10
20
drift time residuals [ns]
OT-t0 calibration: at work
Δt 0[ns]
•  Executed every fill, update expected every few weeks •  Automa3c evalua3on and update of the constants worked from day one!!! •  Added for monitoring purpose the t0-­‐calibra3on per chip (4-­‐each module). 0.4
0.3
0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5
t new
Applied
0
t new
Not applied
0
LHCb OT Preliminary
thr = 0.1ns
thr = 0.04 ns
Commissioning
0
Early Measurements
200
Stable running
400
600
Calibration number [a.u.]
52 SGll a difficult challenge: •  Managing to do the full reconstruc3on in … achieving offline performance (same or bener than in RunI): DONE! •  Managing to do the full alignment and calibra3on of the detectors in real-­‐3me : DONE! •  Being able to select efficiently many signals already at trigger level Barbara Storaci, University of Zurich 53 Selection power at trigger level
25000
20000
LHCb
Preliminary
s = 8 TeV Data
15000
Trigger + PID + 3ghter PID 10000
104
103
LHCb
Preliminary
s = 8 TeV Data
Trigger + PID 102
10
5000
0
1800
Candidates/MeV/c2
Candidates/MeV/c2
•  Same reconstruc3on as offline and complete alignment and PID calibra3on allows to apply a 3ghter selec3on on kinema3cs quan33es •  RICH calibra3on allows to use hadron par3cle iden3fica3on in selec3on, e.g. boost efficiency for Cabibbo suppressed decays while keeping the rate low by pre-­‐scaling the Cabibbo favored counterpart 1850
1900
2
1
1800
mπππ [MeV/c ]
Barbara Storaci, University of Zurich 1850
1900
mKKK [MeV/c2]
54 SGll a difficult challenge: •  Managing to do the full reconstruc3on in … achieving offline performance (same or bener than in RunI): DONE! •  Managing to do the full alignment and calibra3on of the detectors in real-­‐3me : DONE! •  Being able to select efficiently many signals already at trigger level : DONE! Ready to do physics from the HLT output! Barbara Storaci, University of Zurich 55 The TURBO Stream
•  Store HLT candidate informa3on •  Remove detector raw data •  Save ~90% of space •  Idea for analysis with large yields: possible to reduce the pre-­‐scaling of all the channels that were trigger output rate constraint -­‐> more physics possible! •  ~24h turn-­‐around from data taking to analysis •  Real challenge relying on the good opera3onal performance: no 2nd chance of reprocessing to fix problems Barbara Storaci, University of Zurich 56 The TURBO Stream at work: the early measurements
More m
the talk aterial on Spradlin of Patrick 7 th Octo Wednesday ber! Used for early measurement cross-­‐sec3on Results presented 1 week amer their acquisiGon! Example J/ψ cross-­‐sec3on ArXiv 1509.00771 accepted by JHEP –  The trigger found 106 J/ψ-­‐>μ+μ-­‐ in 3.03 ± 0.12 pb-­‐1 with J/ψ pT<14 GeV/c and 2<y<4.5 –  No offline processing: mass resolu3on of ~12MeV/c2 compa3ble with RunI data Barbara Storaci, University of Zurich 57 Luminosity measurement
•  In LHCb two techniques used to measure the luminosity: –  TradiGonal Van der Meer scans: with LHC moving the beams with respect to each other in small steps –  Beam Gas Imaging Method with SMOG to reconstruct beam shape at the interac3on point •  The combina3on of the two techniques allowed in RunI to achieve the best precise measurement of the luminosity at LHC with 1.1% of uncertainty [ JINST 9 (2014) 12, P12005 ] •  Only beam gas imaging method available for the EM 3mescale Only HLT1 is used for this •  Delivered a luminosity measurement with 3.8% of uncertainty in few weeks (measurement 10th measurement: the new HLT split of June, first results presented at EPS the 22th of allowed to write very fast data to July) disk increasing the precision of the •  Further analysis to combine the two methods measurement (~1%) ongoing Barbara Storaci, University of Zurich 58 Conclusions
•  LHCb is the first HEP experiment with a full calibraGon, alignment and reconstrucGon done in real-­‐Gme •  Provided new calibra3on and alignment constants for each run or fill in few minutes •  Possible to monitor in few hours quan33es that in Run I were monitored only few 3mes in a year (like during TS) •  The Turbo Stream allowed to: –  Increase the physics reach by saving more data in less space (important for high branching ra3o channels) –  Physics analysis doable ~24h ajer the data taking •  First measurement of the luminosity at 13TeV with only 3.8% of uncertainty (thanks to the Beam Gas Imaging Method ) in few weeks from the data taking! •  First cross-­‐sec3on measurements presented at summer conferences one week ajer the data taking Barbara Storaci, University of Zurich 59 Backup
Barbara Storaci, University of Zurich 60 The LHCb experiment
•  LHCb is the dedicated heavy flavor physics experiment at LHC •  Its primary goal is to look for indirect evidence of new physics in CP-­‐viola3on and rare decays of beauty and charms hadrons •  This requires: LHCb
s = 8 TeV
10000
8000
6000
4000
candidates / (0.1 ps)
Candidates / (25 MeV/c2)
1.  Excellent tracking (momentum, impact parameters and primary vertex resolu3on) 2.  Excellent decay 3me resolu3on 3.  Excellent par3cle iden3fica3on Tagged mixed
Tagged unmixed
400
Fit mixed
Fit unmixed
200
LHCb
2000
0
9000
10000
11000
m( µ+µ− ) [MeV/c2]
JHEP 06 (2013) 064 0
0
1
2
3
4
decay time [ps]
New J.Phys. 15 (2013) 053021 Barbara Storaci, University of Zurich Eur. Phys. J. C (2013) 73 61 Trigger in RunII
•  Events from lower trigger levels can be buffered on disk while performing real-­‐
Gme alignment and calibraGon •  Last trigger level uses the same reconstrucGon as offline •  Same alignment and calibraGon constants used by the trigger and the offline reconstruc3on •  Some analysis performed directly on the trigger output Barbara Storaci, University of Zurich 63 RICH calibration: logic
•  RICH automa3c calibra3on consists of calibra3ng: –  RICH refrac3ve index: •  Depends on the gas mixture, temperature and pressure •  Determined by fitng the difference between reconstructed and expected Cherenkov angle –  Hybrid photon detector (HPD) images •  Affected by the magne3c and electric fields distor3ons •  Anode images cleaned and Sobel filter is used to detect the edges Fully automa3c procedure, evaluated run by run •  In case a too short run, the same constants of the previous run are used •  Calibra3on parameters are monitored by PVSS trend plots Barbara Storaci, University of Zurich 64 Optimization of the reconstruction chain
•  In order to maintain full reconstruc3on efficiency, while keeping within a strict 3ming budget many improvements were needed: –  Op3miza3on of the code (e.g. vectoriza3on) –  Changes to the reconstruc3on chain and op3miza3on of the panern algorithm Example •  Improved sequence forming VeloTT tracks as an intermediate stage •  Availability of a momentum es3mate to pre-­‐select tracks at an early stage •  Charge es3mate allows greatly reduced search windows downstream of the magnet •  Factor of 4 less ghost rate •  Factor of 3 reducGon in execuGon Gme •  Tracks with a pT>500MeV already available at HLT1 level (in Run I pT>1.3 GeV) without any IP requirements needed. VeloTT algorithm
Barbara Storaci, University of Zurich 66 Calorimeter and Muon systems performance
Electron idenGficaGon efficiency >91% with a hadron mis-­‐idenficaGon fracGon of few % p>10 GeV/c (intermediate selecGon) LHCb Muon efficiency >95% for almost all the pT bins in the full momentum range Int.J.Mod.Phys. A30 (2015) 1530022 Barbara Storaci, University of Zurich 67 The PID performances
Capability to disGnguish parGcles (especially π-­‐K) in a the momentum range 2-­‐100 GeV/c) Kaon idenGficaGon efficiency ~95% with a pion mis-­‐idenficaGon fracGon of ~10% over the full 2-­‐100GeV/c momentum range (Gght selecGon) Eur.Phys.J. C73 (2013) 2431 Barbara Storaci, University of Zurich 68 Server nodes
Barbara Storaci, University of Zurich 69 Gain for b-physics
•  Ra3o of RunII over RunI for HLT2/HLT1 efficiencies. Note that the denominator is reconstruc3ble with pT(B)>2 GeV, τ(B)>0.2ps Barbara Storaci, University of Zurich 70 Tracking efficiency from data
JINST 10 (2015) P02007 Barbara Storaci, University of Zurich 71 VeloTT speedup
Barbara Storaci, University of Zurich 72