atlas simulation/reconstruction software

ATLAS Simulation/Reconstruction Software

Reported by Jim Shank, work done by Reported by Jim Shank, work done by

most US Institutes.most US Institutes.

DOE/NSF review of LHC Software and Computing ProjectsDOE/NSF review of LHC Software and Computing Projects

FermilabFermilab

27-30 November, 200127-30 November, 2001

28 Nov., 2001.28 Nov., 2001.J. Shank ATLAS Simulation/Recon. SW.J. Shank ATLAS Simulation/Recon. SW. 2

Outline

Activities in all systems (mostly by physicists):Activities in all systems (mostly by physicists):

Pixels, TRT, EM Cal, Tile Cal, Muons, Trigger

Well integrated into overall ATLAS computing effort.Well integrated into overall ATLAS computing effort. In particular, the US core efforts on Athena and DB.

Review of recent activity, by sub-systemReview of recent activity, by sub-system

Future work: Data Challenges.Future work: Data Challenges.


ATLAS Subsystem/Task Matrix

Offline Offline

CoordinatorCoordinator

ReconstructionReconstruction SimulationSimulation DatabaseDatabase

ChairChair N. McCubbinN. McCubbin D. RousseauD. Rousseau K. AmakoK. Amako D. MalonD. Malon

Inner DetectorInner Detector D. BarberisD. Barberis D. RousseauD. Rousseau F. LuehringF. Luehring S. BentvelsenS. Bentvelsen

Liquid ArgonLiquid Argon J. CollotJ. Collot S. RajagopalanS. Rajagopalan M. LeltchoukM. Leltchouk R. SobieR. Sobie

Tile CalorimeterTile Calorimeter A. SolodkovA. Solodkov F. MerrittF. Merritt A. SolodkovA. Solodkov T. LeCompteT. LeCompte

MuonMuon To be namedTo be named J.F. LaporteJ.F. Laporte A. RimoldiA. Rimoldi S. GoldfarbS. Goldfarb

LVL 2 Trigger/ LVL 2 Trigger/

Trigger DAQTrigger DAQ

S. GeorgeS. George S. TapproggeS. Tapprogge M. WielersM. Wielers H. P. BeckH. P. Beck

Event FilterEvent Filter V. VercesiV. Vercesi F. TouchardF. Touchard

Other US roles: D. Quarrie (LBNL), Chief Architect; P. Nevski (BNL), Geant3 simu Other US roles: D. Quarrie (LBNL), Chief Architect; P. Nevski (BNL), Geant3 simu coord; H. Ma (BNL), Raw data coord; C.Tull (LBNL), Eurogrid WP8 liaisoncoord; H. Ma (BNL), Raw data coord; C.Tull (LBNL), Eurogrid WP8 liaison


Subdetector SW Activities Summary

Performance/design studiesPerformance/design studies

G3 based simulationG3 based simulation

Test beam Test beam

Athena integration Athena integration

Reconstruction development in CReconstruction development in C++ ++

G4 based simulation developmentG4 based simulation development

G4 physics validation G4 physics validation

XML based detector description XML based detector description

Database Database

Conditions DB

Trigger/DAQTrigger/DAQ


Pixels-Conditions DB (Berkeley)

Goal: To develop general mechanism for retrieving time-dependent alignment Goal: To develop general mechanism for retrieving time-dependent alignment

constants from database and using them in reconstructionconstants from database and using them in reconstruction

Requires additions to Athena infrastructure

Requires extension of existing detector description interface Will prototype using silicon and pixel detectors as the use caseWill prototype using silicon and pixel detectors as the use case Misalignments calculated from numbers stored ”Conditions Database”Misalignments calculated from numbers stored ”Conditions Database”

Delivered through a general ”Time Dependent Conditions Service” in Athena (TCS)

In addition to event store (TES):In addition to event store (TES): Need a detector store (TDeS) Need interface to conditions DB (TCS)

A prototype TDeS coded by C. Leggett and P. Calafiura (a second instance of A prototype TDeS coded by C. Leggett and P. Calafiura (a second instance of the Store-Gate Service without object deletion at the end of each event)the Store-Gate Service without object deletion at the end of each event)

Work in progress…Work in progress…


TRT (F. Luehring/Indiana et al)

Athena Pile-Up Requirements documentation ATL-SOFT-2001Athena Pile-Up Requirements documentation ATL-SOFT-2001

GEANT4 code writingGEANT4 code writing

TRT hit and digitization definitions

TRT GEANT3 code TRT GEANT3 code

current beampipe + geometry updates

TRT material budget

TRT atlsim example on the grid.TRT atlsim example on the grid.

GEANT 3 GEANT 3 GEANT 4 comparisons GEANT 4 comparisons


LAr Simulation (M. Leltchouk/Nevis et al)

LAr simulation coordination: M. Leltchouk/NevisLAr simulation coordination: M. Leltchouk/Nevis

Participation in G4 EM barrel developmentParticipation in G4 EM barrel development

LAr EM calorimeter hits (LArEMHit) were implemented in GEANT4 by

B.Seligman.

The ROOT I/O scheme is used for hit persistency (see The ROOT I/O scheme is used for hit persistency (see

http://www.usatlas.bnl.gov/computing/software/db/LArRoot2.htmlhttp://www.usatlas.bnl.gov/computing/software/db/LArRoot2.html

http://www.usatlas.bnl.gov/computing/software/db/rootio.htmlhttp://www.usatlas.bnl.gov/computing/software/db/rootio.html ) )

Comparisons of GEANT4 Simulations with Testbeam Data and Comparisons of GEANT4 Simulations with Testbeam Data and

GEANT3 for the ATLAS Liquid Argon Calorimeter has been presented GEANT3 for the ATLAS Liquid Argon Calorimeter has been presented

on CHEP2001on CHEP2001


GEANT 4 LAr Simulation


FCal1 Testbeam Setup in GEANT4FCal1 Testbeam Setup in GEANT4-counter

TailCatcher

Cryostat

FCal2 Module 0

Iron Shield

HoleVeto

MWPC

FCal1 Module 0

Argon Excluder

VetoWall

Setup aroundCryostat only!

FCal1

FCal2

FCal1 ElectrodePattern


FCal1: GEANT3/4 Comparisons of Energy ResolutionFCal1: GEANT3/4 Comparisons of Energy Resolution

Noise Cut Dependence

Rela

tive E

nerg

y R

eso

luti

on [

%]

Beam Energy [GeV]

Geant4

Geant3

Rela

tive E

nerg

y R

eso

luti

on [

%]

Beam Energy [GeV]

No Noise Cut

Fit to experimental data

GEANT4 high energyresolution problem ??


LAr Reconstruction -- Major Milestones Met

Early design in PASO: Early design in PASO: Jan. 2000Jan. 2000 Migrate to Athena: Migrate to Athena: May 2000May 2000

LAr Reconstruction used as a test bed for early Athena First application software to successfully migrate to Athena

3 working days at LBL3 working days at LBL

First Common Calorimeter InterfacesFirst Common Calorimeter Interfaces Oct. 2000Oct. 2000 QA review of then available components QA review of then available components Dec. 2000Dec. 2000

S. Albrand (Grenoble)

Combined Reconstruction (egamma) Combined Reconstruction (egamma) Jan. 2001Jan. 2001 Process GEANT4 LAr Hits (Root Objects) Process GEANT4 LAr Hits (Root Objects) Mar. 2001Mar. 2001 Lund ReleaseLund Release June 2001June 2001

Establishing most of the reconstruction chain From G3/G4 Hits to Particle Identification


LAr Data Classes

Data Objects proposed/implemented in March 2001Data Objects proposed/implemented in March 2001

J. Collot et. al.

T.

D.

S.

Simulation Hits

DetRespSimulation RawChannels , Digits

CellBuilder

EventFilter/Builder

Electronics Calib

ClusterReconstruction

Hits

Cells

RawChannels , Digits

Digits

RawChannelsCells

Clusters

RawChannels , Digits

TBConvDigits

Test beam Zebra Tapes

HitConvHits

Objy , ROOT/IO


Comparison to ATRECON


Recent Plots using LAr recon. program


LAr Reconstruction Conclusion

A central framework that is evolving to provide robust support

The reconstruction design has been built over this framework

Much of the ‘Fortran’ code has been migrated.

Validation ongoing, but results are promising.

It now paves the way for work in:It now paves the way for work in: optimizing and developing new algorithms

Physics and Detector performance studies


Tile Calorimeter

Tile Calorimeter DB coordination: T. LeCompte/ANLTile Calorimeter DB coordination: T. LeCompte/ANL Tile Cal reconstruction coord: F. Merritt/ChicagoTile Cal reconstruction coord: F. Merritt/Chicago Tile Cal XML Detector Description has been improvedTile Cal XML Detector Description has been improved

Extended barrel completed non-uniform plate spacing included Extended barrel can be easily repositioned w.r.t. the barrel

allows studying the effect of recently introduced gap Geant4 models have been built from both XML and "by hand“ G4 vs. test beam comparisons just beginning

TileCal per se reconstruction is largely an issue of calibration (convert TileCal per se reconstruction is largely an issue of calibration (convert ADC counts to Energy) ADC counts to Energy) calibration DB access is a goal for late FY2002

TileCal classes have changed to be more in line with LAr classesTileCal classes have changed to be more in line with LAr classes Jet reconstruction classes have been streamlinedJet reconstruction classes have been streamlined


Tucson JetRec Working Group and Tucson JetRec Working Group and SupportersSupporters

Tucson, Arizona, August 20-22, 2001

Tucson JetRec Working Group and Tucson JetRec Working Group and SupportersSupporters

Tucson, Arizona, August 20-22, 2001

Argonne National Lab: Tom LeCompte

Brookhaven National Laboratory(*): Hong Ma, Srini Rajagopalan

TRIUMF: Monika Wielers

University of Arizona: Peter Loch

University of Chicago: Ed Frank, Ambreesh Gupta, Frank Merritt

(*) by phone

Argonne National Lab: Tom LeCompte

Brookhaven National Laboratory(*): Hong Ma, Srini Rajagopalan

TRIUMF: Monika Wielers

University of Arizona: Peter Loch

University of Chicago: Ed Frank, Ambreesh Gupta, Frank Merritt

(*) by phone


Tasklist for the WorkshopTasklist for the Workshop Come up with an improved JetRec software

design in view of recent suggestions for changes:

definition of basic classes -> review of use cases; establish the algorithm flow; first look at the “navigation problem”

First attempt to set up a working group

structure within the Jet/Etmiss performance group:

work plans, deliverables and commitments; reporting to Jet/Etmiss and Software groups; bi-weekly phone conferences Tuesdays, 17:00 (Geneva

time) -> next October 2, 2001!

Come up with an improved JetRec software design in view of recent suggestions for changes:

definition of basic classes -> review of use cases; establish the algorithm flow; first look at the “navigation problem”

First attempt to set up a working group

structure within the Jet/Etmiss performance group:

work plans, deliverables and commitments; reporting to Jet/Etmiss and Software groups; bi-weekly phone conferences Tuesdays, 17:00 (Geneva

time) -> next October 2, 2001!


Algorithmic FlowAlgorithmic Flow

T DS

CaloCellPr ot oJ et Builder

Cal

oCel

ls

Cal

oPro

toJ

ets

Example!! There is NO restriction to Calorimeter Reconstruction Objects or any other specific type in general!

Example!! There is NO restriction to Calorimeter Reconstruction Objects or any other specific type in general!

Kt J et F inder

Jet

s

Jet

s

Kt J et Reconst r uct or

Jet

s

Jet

s

CaloPr ot oJ et Cont ainer

Kt Pr eClust er Builder

Prot

oJet

s

Jet

s

Kt Pr eClust er Cont ainer

Kt Candidat eJ et Cont ainer

Kt J et Cont ainer


Muon Spectrometer

Boston U (J.Shank), U Michigan, Harvard, BNL + CERN, Italy,..Boston U (J.Shank), U Michigan, Harvard, BNL + CERN, Italy,..

Current activity:Current activity: Muon database and detector description

Muon DB coordination: S. Goldfarb/UM

XML detector description: MDTs, RPCs, TGCs implemented; full chain to Geant4 implemented

Geometry ID scheme for all subsystems defined and documented

OO muon reconstruction (Moore) development Integrated into Athena; in repository; in early development

Limited reconstruction in the barrel

Simulation for detector layout optimization

Near term goals:Near term goals: Extend Moore to barrel, update to emerging reconstruction data model.

Trigger TDR studies: L1->L2 rejection, efficiencies

Calibration DB, trigger DB, ongoing detector description work


Physics Performance Comparasion CSC Doublets vs. SingletEndcap Muon System Staging Study B. Zhou & D. Levin, U. of Michigan

Final stateFinal state Recon. Efficiency Recon. Efficiency

degradationdegradation ResolutionResolutiondegradationdegradation

Non-Gaussian tailNon-Gaussian taildegradationdegradation

1 muon1 muon

< 3%< 3% for Pt range for Pt range

(20 – 500 GeV)(20 – 500 GeV)

< 2%< 2%DDP/P degradationP/P degradation(20 – 500 GeV)(20 – 500 GeV)

< 10% for 500 GeV< 10% for 500 GeVmuonsmuons

2 muons2 muons300 GeV 300 GeV

A -> 2 A -> 2 muonsmuons

~ 4% more ~ 4% more events lossevents loss

Mass resolution Mass resolution changed from changed from 3.6% - 4.0%3.6% - 4.0%

< 2% non-Gaussian < 2% non-Gaussian tail increasetail increase

4 muons4 muons150 GeV 150 GeV

H -> 4 H -> 4 muonsmuons

~ 5% more ~ 5% more events lossevents loss

No significant No significant change in mass change in mass

resolutionresolution

No significant non-No significant non-Gaussian tail Gaussian tail

increaseincrease

• US DoE/NSF Lehmann Review recommend the US ATLAS Muon Team build 50% of the CSC chambers at the initial phase of the LHC.• Physics studies used single muons , and double and four muon final states from low mass Higgs decays.

Conclusion: the US CSC muon staging plan has not shown significant impact on The low mass Higgs detections at the Day 1 of the LHC physics run.June, 2001: ATLAS management approved US staging plan.


Investigation of Alignment Criteria in EndCap MDTsDaniel Levin – University of Michigan

Impact on Efficiency and Resolution due to Uncertainties in chamber surveying, placement & orientation

Resolution @ 100 Gev

3

4

5

6

7

8

9

10

0 0.5 1 1.5

mrad Max

dp/p

Green: Rotation about beam

T S

Z

Criterion: Alignment tolerance should be <0.3 mrad

Efficiency

T Axis Misalignment (mrad)


ATLAS Muon Database Contributions (S. Goldfarb)

Overall CoordinationOverall Coordination Management of MuonSpectrometer packages for Event and Detector Description

Reduction of cross-package software dependencies, porting to CMT New packages for Objectivity DDL

Planning document for Detector Description development ATL-COM-MUON-2001-021

Event Model DevelopmentEvent Model Development MuonEvent

Completion of transient G3 hit, container classes for MDT, RPC, TGC Completion of persistent (objectivity) digit, container classes, schema for MDT, RPC, TGC

New Muon Event Model Commencement of discussions with BNL defining project for Muon Spectrometer Coordination with SCT/TRT community

Detector Description DevelopmentDetector Description Development MuonDetDescr

Completion of transient detector description classes for TGC Completion of persistent (objectivity) detector description classes, schema for MDT

MuonAGDD Evaluations of MDT, RPC, TGC descriptions for GEANT4 simulation Development of “compact” syntax definitions for MDT, RPC, TGC and Barrel Toroid Completion of XML description, expansion interface for MDT, Barrel Toroid

HEPDD (http://doc.cern.ch/age?a01380)HEPDD (http://doc.cern.ch/age?a01380) Hosted, Chaired second annual workshop on Detector Description for HEP at CERN


ATLAS Muon Database Contributions

Descriptions of Barrel Toroid (left)

and H8 test beam geometry (below).

Both geometries were generated using

compact AGDD syntax and both were

developed by REU summer students,

under the supervision of S. Goldfarb.


ATLAS Muon Database Planning Data Challenge 0Data Challenge 0

Persistent (objectivity) detector description classes, schema for RPC, TGC Data Challenge 1Data Challenge 1

Access to Geometry Version O in Athena from AMDB (Naples + SG) General Development to Event ModelGeneral Development to Event Model

MuonEvent New packages for technology-dependent software Modifications necessary for new geometry implementation

New Muon Event Model Initial implementation of Muon Digit Container and Identifier Classes (BNL) Implementation of new identifier scheme (BNL + SG)

General Development to Detector Description General Development to Detector Description (These plans detailed in document ATL-COM-MUON-2001-021) MuonDetDescr

Completion and testing of objectivity persistency New AGDD_DetDescrSource classes to interface MuonDetDescr with AGDD

MuonAGDD Completion of syntax, XML descriptions, interfaces for RPC, TGC, CSC, inert

material Extensive testing, evaluation of AGDD with G4 Simulation, Moore, Muonbox

HEPDD--Plan to Host/Chair Third Annual Workshop on Detector Description for HEPHEPDD--Plan to Host/Chair Third Annual Workshop on Detector Description for HEP


Offline Muon Reconstruction (Moore)

MMuon uon OObject bject OOriented riented ReReconstruction (Moore).construction (Moore).

Runs in the Athena Framework using the ATLAS CMTRuns in the Athena Framework using the ATLAS CMT

StrategyStrategy Base algorithms on trigger simulation:

Make roads from trigger chambers MDT Pattern recognition added (see next slides) Fitting from iPat

Graphics currently using GraXML and ATLANTIS


Pattern Recognition: Track Finding

x, y plane

z, y plane

Inner station

Outer station

Middle station


EfficiencyE

ffici

en

cy (

%)

PT (GeV)

Muonbox

MOORE

620 100 300 1000

A Muon track consists ofhits from at least 2 stationsand is successfully fitted.

The efficiency is normalizedto all events with the generated muon within ||<1at the event vertex


ResolutionP

T-r

eso

lutio

n (%

)

PT (Gev)

6 20 100

300

1000The resolution is defined as the of the gaussian fit to the

PTrec/PT

gen distribution


Pull of 1/PT

PT = 1000 Gev PT = 6 Gev

σ = 1.1941 σ = 6.176

The error on 1\PT pull is due to the material

pull = difference between the reconstructed and true values normalised to the

error on the reconstructed value.


Moore Plans

Release the code documentedRelease the code documented

Extend Moore in the End-cap regionsExtend Moore in the End-cap regions

Look into using the Level-1 simulation code directlyLook into using the Level-1 simulation code directly

Need to get the material description Need to get the material description

Plan to use Cobra fittingPlan to use Cobra fitting

Exploring Graphics with Atlantis and continue with GraXMLExploring Graphics with Atlantis and continue with GraXML

Implement the current O-Layout Implement the current O-Layout

Participate in the Data Challenge (1)Participate in the Data Challenge (1)


Trigger/DAQ Offline Software

The ATLAS High Level Trigger (HLT) is mostly a software triggerThe ATLAS High Level Trigger (HLT) is mostly a software trigger

LVL2: Optimized algorithms and simple trigger menus

Event Filter: Offline-like algorithms, full event, and latest calibrations

The The LVL1 triggerLVL1 trigger is a hardware trigger and needs special simulation in offline is a hardware trigger and needs special simulation in offline

TDAQ software is similar to other detector software in terms of offline TDAQ software is similar to other detector software in terms of offline

requirements and applicationsrequirements and applications

Full simulation is used in design and optimization of TDAQ system

Offline software is used to monitor performance (rates, efficiency, single

component performance)

However, very stringent QC needed; “mission criticality”


T/DAQ Offline Software: Status

LVL1 simulation exists in Athena for e/LVL1 simulation exists in Athena for e/// trigger trigger

Recently, most effort has been in the design of the HLT framework. Recently, most effort has been in the design of the HLT framework.

Main requirement in design:Main requirement in design: Use the same software in online and offline environments

Also plan to have similar framework for LVL2 and EFAlso plan to have similar framework for LVL2 and EF Possibly sharing of (some) services and algorithms

Presently evaluating Athena for use as EF frameworkPresently evaluating Athena for use as EF framework If OK for EF, then consider use at LVL2

First cycle of design recently finished; now implementing first prototypeFirst cycle of design recently finished; now implementing first prototype Aim for vertical slice prototype for Spring 2002 Exploitation for HLT/DAQ/DCS TDR in late 2002


HLT Offline Software: Design

High Level Design stage High Level Design stage

is finishedis finished

Aim is to use same Aim is to use same

design for LVL2 and EFdesign for LVL2 and EF

System factorized in System factorized in

work areaswork areas

Steering

Algorithms

Data Manager

Event Data Model

Interactions needed (and Interactions needed (and

ongoing) with offline and ongoing) with offline and

architecture groupsarchitecture groups


Validation of Athena for HLT Use

The ATLAS EF will use selection and classification algorithms derived The ATLAS EF will use selection and classification algorithms derived

from the offline suitefrom the offline suite

Offline software performance therefore has a direct impact on EF farm Offline software performance therefore has a direct impact on EF farm

size and costsize and cost

The HLT community has started “validation studies” (detailed The HLT community has started “validation studies” (detailed

benchmarking) of Athena, offline algorithms, and event modelbenchmarking) of Athena, offline algorithms, and event model

The aim is to set metrics for monitoring trends in software performanceThe aim is to set metrics for monitoring trends in software performance

It is clear that the software is presently far from adequateIt is clear that the software is presently far from adequate Not fair to judge during development phase But benchmarking can (and has) helped spur improvements Feedback during monthly meetings with A-team and regular interactions with

developers

Software performance is also important for offline – hope that offline Software performance is also important for offline – hope that offline

community will continue this workcommunity will continue this work


Summary

New ATLAS framework, Athena, enthusiastically embraced by broad New ATLAS framework, Athena, enthusiastically embraced by broad

spectrum of sub-system community.spectrum of sub-system community.

Many US physicists active in CMany US physicists active in C++++ code development code development

Well integrated into overall ATLAS software effort

Schedule: Schedule:

DC 0 12/01 Should have full OO sw ready.

Still some Fortran wrapping (muons)

DC1 02/02 Large scale simulation/reconstruction.

Some with GEANT4

Objectivity and Root IO.

atlas simulation/reconstruction software

Documents

shank atlas simulationrecon

jim shank

atlas liquid argon calorimeter

overall atlas computing

conditions databasedelivered

testbeam data

c g4

geant3 simu coord