Il Trigger di Alto Livello di CMS

1
Il Trigger di Alto Livello di CMS

• N. Amapane CERN
• Workshop su Monte Carlo, la Fisica e le
  simulazioni a LHC
• Frascati, 25 Ottobre 2006

2
The Compact Muon Solenoid

CALORIMETERS

ECAL
Scintillating PbWO4
HCAL
Plastic scintillator

Crystals

brass


sandwich
IRON YOKE
TRACKER

MUON
ENDCAPS
MUON BARREL
Silicon Microstrips
Pixels




3
LHC Event Rates
s rate _at_ nominal LHC
luminosity
Pile-up
Machine Rate 40 MHz
On-line trigger selection Select 14x105 Decide
every 25 ns!
Acceptable storage rate 100 Hz
Off-line analysis
Signals
Particle mass (GeV/c2)
4
Trigger Architecture

• Start from 40 MHz ? Decision every 25 ns
• Too small even to read raw data
• Selection in multiple levels, each taking a
  decision using only part of the available data
• The first level (L1) is only feasible with
  dedicated, synchronous (clock driven) hardware

40 MHz
100 kHz

• CMS choice All further selection in a single
  phisical step (HLT)
• Build full events and analyze them as in
  offline
• Invest in networking (rather than in dedicated L2
  hardware)

100 GB/s!!
100 Hz
5
Level-1 Trigger

• Custom programmable processors
• To minimise latency
• Synchronous decision every 25 ns
• delayed by 3.2 ms 128 BX (Max depth of
  pipeline memories)
• Max output ? max DAQ input
• Design 100 kHz at startup 50 kHz
• Only m detectors and calorimeters
• e/g, m, t jets, jets, ETmiss, SET

• Selection by the Global Trigger
• 128 simultaneous, programmable algorithms, each
  allowing
• Thresholds on single and multiple objects of
  different type
• Correlations, topological conditions
• Prescaling

6
Trigger detectors

• ECAL up to hlt3
• HCAL hlt 3 (HB, HE) 3lthlt5.191 (HF)
• Muon (DT, CSC, RPC) hlt2.4
• But trigger electronics only up nlt2.1

7
L1 Trigger Table

• For L 2x1033 cm-2s-1
• (CMS Physics TDR v.2)
• Assume 50 KHz DAQ available at low luminosity
  factor 3 safety

8
DAQ
L1
Event building
Modular, 8 slices 4 to be installed at startup
HLT farm (O(2000 CPU)
9
CMS HLT

• Run on farm of commercial CPUs a single
  processor analyzes one event at a time and comes
  up with a decision
• Has access to full granularity information
• Freedom to implement sophisticated reconstruction
  algorithms, complex selection requirements,
  exclusive triggers

• Constraints
• CPU time (Cost of filter farm)
• Reject events ASAP set up internal logical
  selection steps
• L2 muon calorimeter only
• L3 use full information including tracking
• Must be able to measure efficiency from data
• Use inclusive selction whenever possible
• Single/double object above pT/ET, etc.
• Define HLT selection paths from the L1
• Keep output rate limited (obvious)

10
Example Muon HLT
Integral rate (L? 1034 cm-2s-1)
c,b
Rate (Hz)
p?/K?
W
Z/g
100 Hz
KL
t
Threshold on generated pT (GeV/c)

• Key is to achieve the best pT resolution (and
  suppress non-prompt muons and b,c decays)

11
HLT Muon Reconstruction

• Level-2 confirm L1 refitting hits in the muon
  chambers with full granularity
• Regional reconstruction seeded by L1 muons
• Kalman filtering iterative technique
• pT resolution 10 to 16 depending on h (muons
  from W decays)
• Level-3 Inclusion of Tracker Hits
• Regional tracker reconstruction seeded by L2
  muons
• pT resolution achieve full CMS resolution of 1
  to 1.7 depending on h (muons from W decays)
• Isolation in calorimeters (at L2) and tracker
  (L3) to suppress b,c decays and non-prompt muons

12
1/pT Resolution
Level-2 Improve L1 barr. ovr. end. 0.17 0.22 0
.20
Level-3 Full resolution
10x scale
13
Single Muon Rates
L? 1034 cm-2s-1
L2,L3 reduce the rate by improving the pT
resolution L2 is justified as it reduces the rate
to allow more time for processing data from the
tracker
100 Hz
14
HLT Reconstruction

• g
• L2 cluster ECAL deposits into superclusters
  and apply ET threshold
• L3 isolation in HCAL and tracker
• e
• L2 common with g
• L2.5 match the supercluster with a track in the
  pixel detector
• L3 isolation in HCAL and tracker, cut on E/p
• Jets
• Iterative cone algorithm in calorimeters energy
  corrections (non-linearity)
• MET
• Vector sum of transverse energy deposit in
  calorimeters, incl. muons
• Tau
• Look for isolated narrow jet, either
• Isolation in ECALpixel
• Isolation in the tracker
• B-tagging
• L2.5 impact parameter with pixel track stubs
• L3 with regional track reconstruction

15
Setting trigger tables

• HLT trigger paths start from corresponding L1
  paths
• Tresholds are set distributing bandwidth to the
  various paths in order to maximize efficiencies
• There can be significant overlaps
• Iterative process
• Thresholds (and streams) will change with
  luminosity
• And according to the physics of interest at the
  time of operation
• Reference 2x1033 cm-2 s-1
• Evolution of selection with luminosity is a
  delicate issue, up to now studied in detail only
  for jet (with prescales)

16
HLT Trigger Table

• L 2x1033 cm-2s-1
• (CMS Physics TDR v.2)

contd
17
HLT Trigger Table (cont).

• L 2x1033 cm-2s-1
• (CMS Physics TDR v.2)

120 Hz
18
Some HLT Efficiencies

• At low luminosity, relative to events in detector
  acceptance
• W ?en 68
• W ?mn 69
• Z ?mm 92
• Z ?ee 90
• tt ?mX 72
• H(115 GeV)?gg 77
• H(150) ?ZZ?4m 98
• H(120) ?ZZ?4e 90
• A/H(200 GeV)?2t 45
• H(200-400)?tn 58

19
Triggers and offline analysis

• The HLT selection can have an impact on analysis
• May reduce signal efficiency and phase-space
• Unless off-line selection is tighter than HLT
• Simulation of the HLT selection is a part of
  analysis!
• Specific exclusive triggers can be implemented
  for channels where the default trigger tables are
  not enough, but
• How much the selection costs in term of rate and
  CPU?
• Is it possible to understand the selection
  efficiency from the data?

20
Conclusions

• Trigger at LHC is an integral part of the event
  selection
• CMS uses a single physical step after L1, to
  achieve a rejection factor of 1000
• HLT algorithms have the full event data available
  and no limitation on complexity, except for CPU
  time
• Inclusive triggers based on the presence on one
  or more objects above pT/ET thresholds are
  normally sufficient to get good efficiency on
  most signal
• More sophisticated selections are possible if
  necessary

21
References

• CMS DAQ/HLT TDR, 2002, CERN-LHCC-2002-026
• Full study of HLT rates, timing, benchmark signal
  efficiencies
• CMS Physics TDR Volume 1 (2006),
  CERN-LHCC-2006-001
• Detector performance, reconstruction
• CMS Physics TDR Volume 2 (2006),
  CERN-LHCC-2006-021,
• Update of HLT rates and trigger tables (Appendix
  E)