The ATLAS High Level Trigger

Véronique Boisvert

CERN

On behalf of the ATLAS Trigger/DAQ High Level Trigger Group

Université de Montréal-McGill Seminar August 18th 2003

Rockefeller Center NY, USA

V. Boisvert

Outline Physics Motivation Selection Strategies ATLAS detector LHC environment Trigger Architectures High Level Trigger (HLT) Selection Software Measurements Conclusions

V. Boisvert

The Big Questions

Is there unification of all forces?

What breaks it?

What breaks EW Symmetry?What is the origin of mass?

What is the physics beyondThe SM? New particles?

New interactions?

Flavor Puzzles:Can we understand the masses

And mixing of fermions. Where does CP come from?

Are there more forces? Particles? Symmetries?

Explain the masses of The p and e, and the Relative strengths of

The fundamental forces

Do we understand theStructure and fate of

The universe?

Are there extraDimensions? What is the structure of spacetime?

What is the right descriptionOf gravity and where does itBecome relevant for particle

Physics?

VLHC100TeV

pp

0.5-1.0 TeVe+e-

Collider

MuCollider

Nu Factory

High LuminosityZ Factory

B,K,tau/charmFactory

Tevatron2TeV

pp

ParticleAstrophysics

14 TeVPp

LHC

Can we explain the universe?Why is it matter dominated?

Cosmological Constant?Dark Matter Problem?

Adapted from fig. From P. Drell, published in Physics Today Jan 2001

V. Boisvert

Some Answers from the LHC Electroweak symmetry breaking Precise Standard Model measurements B physics Physics beyond the Standard Model:

SUSY Exotics

The unknown!

V. Boisvert

Electroweak Symmetry Breaking SM Higgs:

114.4GeV < mH < 1TeV

LHC Higgs production and cross-sections

Higgs decays: Fully hadronic:

Large QCD background Gold plated modes:

H Signature: pT >= 50GeV/c

~6 for mH=120GeV, 30 fb-1

V. Boisvert

Electroweak Symmetry Breaking

Gold plated modes: H ZZ(*) 4l

• Signature: 4 high pT l

• =3-25 (dep. mH), 30fb-1

• Other typical signatures:• tt,bb,ll,ll,lljj

• MSSM Higgs• Typical signatures for H0, h0, A,

H:• ,,,tb

V. Boisvert

Precision Measurements of SM High Luminosity and High E

LHC is the ultimate factory: B, top, W, Z, H, … 1:1013 for Higgs

Deviations from SM Hints of new physics

Precise W mass W jj

• Large QCD background

• W e()• reco. in transverse plane!

V. Boisvert

Precision Measurements of SM Precise W mass

Very dependent on E scale (0.02%)

Built-in calibration system e,, ATLAS, CMS: mW~15MeV

(today ~34MeV) Precise Top mass: tt

t Wb Signatures:• Jets (including b-jets), l, Et

miss

• All channels, ATLAS, CMS: mt~1-2GeV (today ~ 5.1GeV)

• Indirect mH~25%! (today ~50%)

LHC

V. Boisvert

B physics Copious production of B’s:

CP-violation, Bs oscillations, Rare decays, etc.

Bd J/ KS

• Max performance: (sin2)=0.010• Min performance: (sin2)=0.016

• Rare decays• Forward-Backward A: B0

d K*0 +-

• Lowest mass region: enough accuracy to detect New Physics

• Signatures: di-leptons (), semi-exclusive reconstruction

q2/MB2

AFB

V. Boisvert

SuperSymmetry SM is an effective theory:

Gauge coupling unification (families, gravity, etc.) Fine-tuning Hierarchy problem

SUSY: supersymmetric partners s-1/2 Pros:

Elimination of fine-tuning by exact cancellations between partners Quark masses: radiative corrections in SUSY Consistent with string theories (incl. gravity)

Cons: No observation! broken, many free parameters and extensions

If weak-scale SUSY exists the LHC experiments will discover it!

V. Boisvert

SUSY MSSM particle spectrum,

current limits: ml, > 90-100 GeV (LEP) mq,g > 250 GeV (Run 1)

Lightest SUSY Particle (LSP) is 1

0

Cold dark matter candidate Do neutralino reconstruction!

Signature: ETmiss

Decay chains No SM background, 2-body

kinematics Need jets, l, ET

miss

ml

qL~

~

q

±l ±l

~

R

V. Boisvert

Beyond the SM SUSY, Technicolor, Little Higgs, New fermions and

gauge bosons, compositeness,…

Large Extra Dimensions Solves hierarchy problem:

1 fundamental scale: EW scale (TeV) Gravity is weak because propagate in

3+n dimensions Cosmological implications

Constraints from astrophysics Possible explanation for dark matter Etc.

Tests Gravity and String Theory in the lab!

3-branebulk

V. Boisvert

Beyond the SM n2: ADD

Graviton emission Signature: jet() + ET

miss

Randall-Sundrum: n=1 Warped 2 branes (Planck and TeV)

Radion: represents fluctuations of the distance between the 2 branes

Signature: Higgs like Mini black holes!

Gr

r

V. Boisvert

So far…

With a little bit of luck the LHC could completely revolutionize our field!

Highlighted possible signatures Other constraints on the trigger architecture?

V. Boisvert

The LHC at CERN

From: P. Sphicas 2003

V. Boisvert

The LHC environment Interaction rate: L x (pp) = 1034cm-2 s-1 x 70mb =

107mb-1 Hz x 70mb = 7x108Hz! ~3600 bunches in LHC

Length of tunnel is 27Km Time between bunches: 25ns! (40MHz bunch x rate)

V. Boisvert

The LHC environment Interactions per crossing: ~23!

Minimum bias events overlap each event of interest We have “pile-up”

“In-time”: particles from same crossing but different pp interaction

“Out-of-time”: left-over signals from previous crossings

Need bunch crossing identification

V. Boisvert

Time of flight…

Weight: 7000 t 44 m

22 m

~108 channels (~2 MB/event)

V. Boisvert

pp collisions at high luminosity

HZZ 4

V. Boisvert

T/DAQ challenges efficient signal selection and excellent

background rejection Interaction rate: 7x108 Hz

Store data at 100 Hz Bunch crossing rate: 40MHz

Out of time Pile-up Synchronization over detectors

High number of channels at high occupancy It’s online!!

If event is not selected it’s lost forever!

V. Boisvert

Selection Strategies 2 main guiding principles:

Inclusive selection Mostly 1 or 2 objects (electron, muon, photon, jet, b-

tagged jet, tau, ETmiss, ET)

High pT : > O(10GeV/c)

Worry about: Low mass objects (eg B physics)

Exclusive selection, topology, etc. Biases in selection

Use complementary selections

V. Boisvert

Selection Strategies

ObjectObject Examples of physics coverageExamples of physics coverage NomenclatureNomenclature

ElectronsHiggs (SM, MSSM), new gauge

bosons, extra dimensions, SUSY, W, top

e25i, 2e15ie25i, 2e15i

PhotonsHiggs (SM, MSSM), extra

dimensions, SUSY60i, 260i, 220i20i

MuonsHiggs (SM, MSSM), new gauge

bosons, extra dimensions, SUSY, W, top

20, 220, 21010

Jets SUSY, compositeness, resonances j360, 3j150, 4j100j360, 3j150, 4j100

Jet+missing ET SUSY, leptoquarks j60 + xE60j60 + xE60

Tau+missing ETExtended Higgs models (e.g. MSSM),

SUSY30 + xE4030 + xE40

V. Boisvert

So far…

The LHC environment is brutal to a Trigger DAQ system How to get the job done:

Trigger Architecture

V. Boisvert

The ATLAS Trigger Architecture

40 MHz

75 kHz

~1 kHz

~100 Hz

~1 sec

~10 ms

2.5 s

Rate Latency

Level 1Level 1triggertrigger

Hig

h L

evel Tri

gger

Hig

h L

evel Tri

gger Level 2Level 2

triggertrigger

EventEventFilterFilter

Reg

ion

of

Inte

rest

RoI

V. Boisvert

Introduction: Regions of Interest Typically a

few ROI / event Ex: Pixel 0.2x0.2

~ 92 Modules ~ 332 channels

Only few % of event data required!

V. Boisvert

ATLAS, CMS vs Other detectors

V. Boisvert

ATLAS vs CMS ATLAS:

Smaller bandwidth But more complex

CMS: Simpler system But very high

bandwidth dependent on

technology

V. Boisvert

So far…

Introduced ATLAS Trigger Architecture Let’s look at the HLT Selection Software

Handle to making the Trigger decision Measurements

V. Boisvert

HLT Selection principles Fast

Early rejection Seeding Data on demand (RoI or whole event)

Modify easily signatures Precise knowledge of detectors and algorithms:

offline community Use offline code in the HLT software

Develop Trigger Alg in offline framework Study boundary between Level 2 and EF Performance studies for physics analysis

V. Boisvert

HLT Selection principles Offline into online: not an easy task!

Requirements of speed and multi-threading on core infrastructure

different steering philosophy: Offline: typically process entire events in a sequential fashion (post

data on a whiteboard) Online: seeded and early rejection

Appointment of a Reconstruction Task Force Look at issues regarding offline-online unification High Level Design (data flow, EDM)

Subdetectors reconstruction Combined reconstruction Analysis preparation reconstruction

General Design principles

V. Boisvert

HLT Design Overview

V. Boisvert

HLT Selection Software

HLTSSW

Steering

ROBDataCollector

DataManager

HLTAlgorithms

Processing Application

EventDataModel

Processing Application

Interface

Dependency

Package

Event Filter

HLT Core Software

HLT Algorithms

Level2

HLT Selection Software

HLT DataFlow Software

HLTSSW at work: 2e30i

V. Boisvert

The Steering

Requirement: Early rejection

Chosen strategy: Seeding mechanism Step wise process

Isolation

pT>30GeV

Cluster shape

trackfinding

Isolation

pT>30GeV

Cluster shape

trackfinding

EM20i EM20i+

e30i e30i +

e30 e30+

e e +

ecand ecand+

Signature

Signature

Signature

Signature

Level1 seed

STEP 1

STEP 4

STEP 3

STEP 2

Steering

V. Boisvert

HLT algorithms: e, selection Level1: selects calorimeter

info over coarse granularity Level2:

1)cluster E, position, shower-shape variables

Refine L1 position: max E (1, 1) Refine (1, 1) with Energy

weigthed average in window 3x7: (c, c)

Parameters to select clusters: Sam. 2: R

shape = E37/E77

Sam. 1: Rshape = E1-E2/E1+E2

Etotal in 3x7 around (1, 1) Ehad in 0.2x0.2 around (c, c)

EM LAr calorimeter

~190,000 channels

For 25GeV: E/E~7%, ~8mrad, r~1.6mm

HLTAlgorithms

V. Boisvert

HLT algorithms: e, selection Level 2:

2) need Track in InDet for el: Pixel, SCT algorithm

Z-finder Hit Filter Group Cleaner Track Fitter

z

Momentum res.: pT/pT ~ 0.1 pT (TeV)

Impact parameters: r< 20m z < 100m

HLTAlgorithms

V. Boisvert

HLT algorithms : e, selection Event Filter:For electrons passing Level 2,

reexamined at EF Use offline reconstruction algorithms Calibrated data for the InnerDetector More tools for reconstruction since full event

Measurements: single el, pT=25GeV/c Fully simulated events, latest software Pile-up for low and high lum Up to date geometry, amount of material, B field

HLTAlgorithms

Trigger Step Rate (Hz) Efficiency (%)Level2 Calo 2114±48 95.9±.3

Level Tracking 59 ±4 88. ±.5Level Matching 37 ± 86.6 ±.6

EF Global 3 ±5 79. ±.7

V. Boisvert

The Data Access

Algorithm RegionSelector

HLT Algorithm

RegionSelector

Trans.EventStore

Data Access

ByteStream

Converter

Data sourceorganized

by ROB

TransientEventStore

region

list DetElem IDs

ROB ID

raw event dataDetElems

list DetElem IDslist DetElem IDs

DetElems

DataManager

V. Boisvert

Collection Number Number of ROBsPixel module 1744 81SCT side of module 8176 256TRT straw layer 19008 256LAr Trigger Tower 7168 768Tile module 256 32Muon MDT chamber 1168 192Muon CSC chamber 32 32Muon RPC chamber 574 32Muon TGC chamber 1584 32

Data access granularity

Preliminary

V. Boisvert

The Event Data Model Raw Data in byte stream format

Level1, Level2, EF results, ROB data Different formats of Raw Data for particular subdetector

RawDataObjects are object representation of Raw Data For InnerDetector the RDOs are skipped for Level2 (data

preparation in converters) Features

Clusters, Tracks, electrons, jets, etc. MCTruth info

For debugging and performance evaluation Trigger Related data

ROI objects, Trigger Type, Trigger Element, Signatures

Offline dependencies!

EventDataModel

V. Boisvert

HLT Selection Software

HLTSSW

Steering

ROBDataCollector

DataManager

HLTAlgorithms

Processing Application

EventDataModel

Processing Application

Interface

Dependency

Package

Event Filter

HLT Core Software

HLT Algorithms

Level2

HLT Selection Software

HLT DataFlow Software

StoreGateAthena/Gaudi

<<import>><<import>>

Offline Architecture &Core Software

Offline Reconstruction

Algorithms

<<import>>

Offline Reconstruction

<<import>>

Offline EventDataModel

V. Boisvert

Timing Measurements

Timing on 2GHz machineLevel2 Calo ~2msLevel2 Tracking ~3ms~EF ~0.5s

Steering

Algorithms

Region Selector

Data Access

1GHz, 3 seeds: 1.2ms

Infrastructure: ~23s :Data access Muons <8 ( )ms GHz

/Lar Tile < ( )ms GHzInnerDetector improvements underway

1GHz, Tile: 0.03ms, Pixel:0.2ms, TRT:1.1ms

V. Boisvert

Measurements Putting it all together in the most realistic

environment: the Level 2 Test bed

Time[ms] Time[ms]

V. Boisvert

Conclusions

The LHC: quite a challenge! The LHC detectors Trigger DAQ systems

Interesting comparisons coming! The ATLAS architecture

RoI mechanism Use of offline code in online environment HLT selection software is adequate and performant

V. Boisvert From: P. Sphicas 2003