ild software for ild detector simulation and optimization · similar detector model lcio helps to...
TRANSCRIPT
ILD
Software for ILDdetector simulation and
optimization
Akiya MiyamotoKEK
12-July-2010DESY Computing Seminar
ILD ContentesILC overviewGLD software tools and optimizationILD Software tools
MokkaReconstruction
GRID in KEK
12 July 2007 Akiya Miyamoto 2DESY Computing Seminar
ILD International Linear Collider : ILCe+e- ColliderEcm: 0.2 ~0.5 TeV 1TeV∫Ldt = 500 fb-1 in 4 years
2004: launched2007: ILC RDR2009: Detector LOI
2012: TDR/DBD
12 July 2007 Akiya Miyamoto 3DESY Computing Seminar
ILD Higgs – Vacuum
Higgs coupling proportional to mass in the SM
Did Higgs break the symmetry ?
Condensed in vacuum ?
12 July 2007 Akiya Miyamoto 4DESY Computing Seminar
ILD Supersymmetry
Fermion Boson
• Do forces unify ?• Do Quarks and Leptons unify ?
• Mass, Coupling
Extending to Higher energy/Early Universe
12 July 2007 Akiya Miyamoto 5DESY Computing Seminar
ILD ILC Reference Design ReportReleased 2007 summer: http://www.linearcollider.org/cms/?pid=1000437
Executive Summary
Physics at the ILC
Accelerator Detectors
RDR: 4 volumes/ 774 pages
12 July 2007 Akiya Miyamoto 6DESY Computing Seminar
ILD Challenge of ILC experimentse+e- Collision:Well defined initial states and relatively clean final states.
Find faint new physics signalsMake precise tests of theory
Many events contains top, W, Z jetsPrecise measurement of jet energy
Calorimeter system inside the coilHighly segmented calorimeter for high resolution
Efficient b/c tagging crucialThin material, strong B field, VTX very close to IP, pixel detectors
Higgs recoil measurements ( e+e- Zh llXh) require good ΔP/PHermetic detector down to very close to beam pipe ( ~ 10 mrad )Detector should be shielded well against beam related backgrounds,low energy e+e- pairs, γγ hadrons, muons,
12 July 2007 Akiya Miyamoto 7DESY Computing Seminar
ILD Detectors for ILC experiments
Tracker: Charged particlesΔpt/pt : 1/10 of LHC
Higgs recoil to Z
Calorimeter:Neutral particlesΔE/E <½ of LEP
good badW/Z separation
WWZ
Z
Vertex Detectorb/c/τ taggingRvtx<1/5 LHC
Improves η(c-tag)
12 July 2007 Akiya Miyamoto 8DESY Computing Seminar
ILD
DESY Computing Seminar
Challenge of SoftwareChallenge
Enrich ILC physics case, taking progresses of HEP ( work with theorists )Show the proposed detector can do physics and meet the ILC goal
Performance targetParticle Flow : Very good jet energy resolution by highly segmented calorimeterVertexing: 2nd/3rd vertex reconstrucionTracking: excellent ΔP/P in ILC condition
Studies based on full simulation and realistic reconstruction are necessary
MokkaMarlin Reconstruction
PandoraPFALCFIVertexTrack Reconstruction
912 July 2007 Akiya Miyamoto
ILD
DESY Computing Seminar
GLD+LDC ILDAt the time of ILC RDR, 4 detector concepts were considered.For LOI submission in 2009, GLD and LDC agrees to merge and formed the ILD group.
In order to define ILD concepts, studied benchmark processeschecked consistency of software toolsoptimized ILD parameters
1012 July 2007 Akiya Miyamoto
LDC:Small cell CAL.Gaseous Tracker4TEuropean based
LDC:Small cell CAL.Gaseous Tracker4TEuropean based
GLD:Small cell CAL.Gaseous Tracker3TAsian based
GLD:Small cell CAL.Gaseous Tracker3TAsian based
ILD
ROOT objects : Event Tree & Configuration
GLD software tools
BeamtestAnalysis
EventReconstruction
DigitizerFinderFitter
DetectorSimulator
QuickSimFullSim
EventGenerator
PythiaCAINStdHep
PhysicsAnalysis
Jetfinder
Link to various tools at http://acfahep.kek.jp/subg/sim/softGLD Software at http://ilcphys.kek.jp/softAll packages are kept in the CVS. Accessible from http://jlccvs.kek.jp/
12 July 2007 Akiya Miyamoto 11DESY Computing Seminar
ILDJSF
Framework: JSF = Root based applicationAll functions based on C++, compiled or through CINTProvides common framework for event generations, detector simulations, analysis, and beam test data analysisUnified framework for interactive and batch job: GUI, event displayData are stored as root objects; root trees, ntuples, etc
development has started since 1999
Release includes other toolsQuickSim, Physsim(event generators)BSGen(generate luminosity spectrum)Analysis utilities…
12 July 2007 Akiya Miyamoto 12DESY Computing Seminar
ILDJupiter/Satellites for Full Simulation Studies
JUPITERJLC Unified
Particle Interactionand
Tracking EmulatoR
IOInput/Outputmodule set
URANUS
LEDA
Monte-Calro Exact hits ToIntermediate Simulated output
Unified Reconstructionand
ANalysis Utility Set
Library Extention for
Data Analysis
METISSatellites
Geant4 basedSimulator
JSF/ROOT basedFramework
MC truth generator Event Reconstruction
Tools for simulation Tools For real data
Jupiter has modular structure for easy installation of sub-detectorsJupiter can run as a standalone job or a module of JSFGeometry parameters are set by an ascII file read-in at run time Special feature to store pre- and post- point of tracks before/after Calorimeter and break points for PFA studies
12 July 2007 Akiya Miyamoto 14DESY Computing Seminar
ILD ILD Optimization Procedure
1512 July 2007 Akiya Miyamoto DESY Computing Seminar
WhizardWhizard PhyssimPhyssim
StdHepStdHep
MOKKAMOKKA JupiterJupiter
LCIOLCIO
MarlinMarlin SattelitesSattelites
LCIOLCIO
DST and AnalysisDST and Analysis
LDC GLDStdHep: Same generator dataLCIO: Common IO format GLDPrim/LDCPrim:
Similar detector model
LCIO helps to collaborative works for detector optimization
Software inter operativity
After ILD optimization, LDC framework was selected as the baseline for LOI studies. No time to really merge GLD and LDC software
ILD Optimization by Benchmark Process
1612 July 2007 Akiya Miyamoto DESY Computing Seminar
0 01 1
0 0 0 02 2 1 1
e e W W W W
e e ZZ
χ χ
χ χ χ χ
+ − + − + −
+ −
→ →
→ →
% % % %
% % % %
Using several detector models, performance to separate W/Z in jet mode have been studied using SUSY processes
by Taikan Suehara
No significant differences are seen
ILD Benchmark : 500 GeV τ pair
1712 July 2007 Akiya Miyamoto DESY Computing Seminar
Only significant difference among detector models found for τ full reconstruction, example in τ− ρ−ντ π−π0ντ
For reconstruction of both g from π0 γγSmaller segmentation (5x5mm2) and larger radius advantageousImpact on physics sensitivity less pronounced
Jupiter
Mokka
Jupiter
Mokka
ILD ILD Design
1812 July 2007 Akiya Miyamoto DESY Computing Seminar
3x Dbl. Layer VTX
Support of BP/VTX/SIT
Forward ComponentBox support option
B=3.5T, RECAL=1.85 m B=3.5T, RECAL=1.85 m
ILD ILD LOI and beyond
1912 July 2007 Akiya Miyamoto DESY Computing Seminar
LOI was submitted March 2009
Validated by IDAG in September 2009
Next step is to develop Detailed Baseline Design
(DBD) by 2012Re-baseline of ILC, working together
with accelerator colleagues
ILD GDE Schedule
2012 July 2007 Akiya Miyamoto DESY Computing Seminar
ILD
DESY Computing Seminar
Software in DBD eraGuidelines for software related studies given by Research Director
“Develop a realistic simulation model of the baseline design, including faults and limitation”“Simulate and study updated benchmark processes including 1 TeV, with background conditions and demonstrate physics capabilities”
ILC re-baselineGDE is updating ILC parameters, taking account R&D progressed since RDRNew parameter will affect beam background conditions and physicsperformance software based studies are necessary
For ILDImplementing GLD goodies to LDC and improve LDC soft to meet requirements for DBD ILDsoftImprove our tools taking into account lessons in LOI era to meet RD’s request
2112 July 2007 Akiya Miyamoto
ILD
DESY Computing Seminar
ILD Software toolsGenerators
Common Stdhep dataWhizard/PhysSim packages
Simulation: MokkaGeant4 application
ReconstructionMarlin FrameworkReconstruction tools as Marlin Processors
Core toolsLCIO : standard for persistency format and event data modelGear, LCCD, CED, …Grid tools and ilcsoft-install
2212 July 2007 Akiya Miyamoto
Digitization
Simulation: MokkaGeant4 application
ILD
DESY Computing Seminar
Mokka simulationCore developped by LLR, using Geant4Geometry data are given by MySQL database
“scalable geometry” has been useful for ILD optimizationMany sub-detector configuration co-exists, even for beam-test detectors.
ILD_00 model ILD_nfairly detailed geometry
Mokka readsILC Common Generator samples in stdhep formatGunieaPig beackground particle data….
Mokka outputsSimCalorimeter, SimTrackerHitsby LCIO
2312 July 2007 Akiya Miyamoto
ILD in Mokka by 3D pdf
ILD
DESY Computing Seminar
VXD in MokkaTwo geometries are available in Mokka
Common to DEPFET, FPCCD, CMOSCryostat is present, butcables and sensors are not addressed.
to be improved for DBD
2412 July 2007 Akiya Miyamoto
Sensor structure
Used for ILD-LOI
ILD
DESY Computing Seminar
Silicon Trackers and TPC
2512 July 2007 Akiya Miyamoto
4 Silicon trackers in ILD: SIT, FTD, ETD, SETCylinders/Disks
strip sensors for DBD
Geometries in Mokka
FTD
SIT
ETD
TPCGas- Ar/CF4/C4H10 Cu, mylar, G10, Air for Field cage and end plate ( equivalent mass )No phi-dependence, but OK. TPC is very uniform device
ILD
DESY Computing Seminar
ECAL
Mixed readout ( ScEcal/SiECAL )will be considered in DBD study.
2612 July 2007 Akiya Miyamoto
ScECAL
ECAL module side view : incl. dead spaces
SiECAL
Two readout options, sharing same structure: Silicon and Scintillator
5x5mm2 Si
SiECAL : baseline for LOI, detailed strcture used for LOI study
ILD
DESY Computing Seminar
HCALAnalog HCALActive: Scintilator
2712 July 2007 Akiya Miyamoto
cylindrical8/16-sided Cross section of1 module
1 layer
Digital HCALActive: RPC
spacer
Float Glass
MylarGraphite
PCB
Elec.
Float glassGraphiteMylar
free space
RPC ( cross view)
Realistic geometry already implemented in MokkaOptimizations :
8/16-sided vs cylindrical & scintillator vs RPCthickness ( # layers ), gaps, tail catchers, absorber materials,
ILD
DESY Computing Seminar
Forward detectors in MokkaConsists of LCAL, BCAL, LHCAL, Beam tube and MasksMokka model ~ CAD model, but CAD model will evolve with time and Mokka model needs to follow
2812 July 2007 Akiya Miyamoto
ILD CAD model
ILD Mokka model
LHCALBCALNew LCAL driver
Tile gap
FEchips
side view of 1 layer
Front view
ILD
DESY Computing Seminar
Muon system, Coil and Yoke Updated version of the Muon system has been prepared.Fairly detailed geometry, waiting integration in the central code
Magnetic fieldFor LOI study,
uniform Solenoid field for physics studyapproximated anti-DID field for background study
For DBDBetter anti-DID field map is necessary for performance study, at least.Uniform field or realistic anti-DID:
– need to consider balance amongcode readiness, CPU penalty, improvements in reality, ……
2912 July 2007 Akiya Miyamoto
Barrel MuonTail catcher in cryostat
ILD
DESY Computing Seminar
Cables/ServicesCables, services, dead materialsfor data out/power in/cooling/gas flow
sub-detector drivers implements their own materials.
To address materials in sub-detectorboundaries, small WG has been setup within ILD for
coordination between sub-detectors/optional detectorsdefining layout and material budgetsImplementation in Mokka will follow
Under new European AIDA framework, new geometry tool kits are indevelopment. New kit will allow consisten geometry treatment among CAD Model, Simulator model, Analysis model. We will be benefitted from this new development
3012 July 2007 Akiya Miyamoto
Mokka: Inner part of ILD
???
ILD
DESY Computing Seminar
Reconstruction ToolsMarlin analysis flow
Digitizer for all sub-detectorsTracking:
LDCTracking for TPC and VTX/Silicon trackers migration to new C++ version in progress
Particle FlowCalorimeter clusteringAssociate track and calorimeter and create PFObjects ( = primaryparticles )
Jet clusteringLCFIVertexing : tag each jets
Output as REC data and DST data by LCIO
3112 July 2007 Akiya Miyamoto
ILD Jet Measurements in ILC Det.Typical feature of ILC events and detector
e+e- Z0 q qbar
ΔE/E
(Δp/
p)
p(GeV/c)/E(GeV)
HD CALΔE/E=50%/√E
Tracker(TK)Δpt/pt=5x10-5pt
EM CALΔE/E=15%/√E
Resolution of a ILC detector
Principle of PFA detectorCharged particles by tracking deviceRemove charged particle signals in CAL ( avoid double counting )
Large bore magnet, Large B-fieldHighly segmented CAL to separate clusters by charged and neutral particles.
Patten reconstruction is a key.
Large bore magnet, Large B-fieldHighly segmented CAL to separate clusters by charged and neutral particles.
Patten reconstruction is a key.
12 July 2007 Akiya Miyamoto 32DESY Computing Seminar
ILD
DESY Computing Seminar
Pandora PFA
3312 July 2007 Akiya Miyamoto
Originally developed by Mark Thomson (U. Cam) as a Marlin Processor.V3.2 was used for ILD LOI. Used by SiD ( thanks to common LCIO format ), achieved ΔE/E ~ 25%/√E for Z pole jetsNow re-organized to PandoraPFANew, as a stand alone package
PFA algorithm
slide by John Marshal (ILD Soft WS)
ILD
DESY Computing Seminar
LCFIVertexingLCFI group developed LCFIVertexing package.Apply algorithm for each jetsIt consists of two parts,
ZVTOP/ZVKIN : Find vertexies from probability of overlapped trajectoriesNeuralNet for tagging and vertexing
Jet with 1 vertex (=IP) – may contain 1 displaced track (D)
> 1 vertecies:– Pt corrected vertex mass is a very good
variable to identify quark floavor.– Other variables (joint-track probability, etc.)
LCFIVertexing also outputvertex charge: powerful discriminator of b and anti-b quarks
Used by both SiD and ILD ( thanks to LCIO )
3412 July 2007 Akiya Miyamoto
Pt corrected vertex mass
ILD
DESY Computing Seminar
LCFIVertexingTypical b/c tag performanceWith neural nets tuned for Z qq eventsVery good performance ( also thanks to the vertex detector placed very close to IP)Issues to be studied
performance with beam backgroundsperformance in multi-jet environments
3512 July 2007 Akiya Miyamoto
Vertex charge of b-jets in ttbar events
Eff. ~ 28% with purity 75% for a b-jet, incl. b B0
LCFIVertexApply neural net analysis for each jets.Jets with no displaced vertex, but >1 misplaced track.
C-jets ( charged D meson)2 misplaced track : b c jetsjoint probability of all track to originate from primary vertex
More than one vertex founddecay length and decay length significancemomentum of tracks assigned to verteciespt corrected vertex massnumber of tracks of non-primary vertexsecondary vertex probabilityjoint probability of all tracks to originate from primary vertex
GRIDGRID is an infrastructure for a large scale international researchesGRID provides
Resources forLarge scale computingLarge scale data storage
International/Inter-regional communication basis
GRID have been used extensively in ILD LOI studies for MC productionsData sharing between Japan – Germany/France/UK
Tohoku Univ.
KEKUniv. of Tsukuba
Nagoya Univ.
Kobe Univ.Hiroshima IT
Network in Japan and GRID• Major HEP projects:
– Belle, J‐PARC, ATLAS ongoing projects– ILC, Belle2 future projects
• Also covering – Material science, bio‐chemistry and so on using
synchrotron light and neutron source– Radiotherapy as technology transfer
• KEK has a role to support university groups in these fields.– including Grid deployment/operation.
Hiroshima Univ. (Alice)U of Tokyo (Atlas)
SINET3
3912 July 2007 Akiya Miyamoto DESY Computing Seminar
http://www.sinet.ad.jp/
Round Trip Time: KEK IHEP/KISTI ~ 100msecFNAL ~ 200msecDESY/IN2P3 ~ 300msec
GRID infrastructures
4012 July 2007 Akiya Miyamoto DESY Computing Seminar
Middle ware gLite NAREGI Gfarm SRB iRODSBelle (Belle2) Using Planning Using UsingAtlas UsingMedical Apps Using Developing PlanningILC Using Planning PlanningJ-PARC Planning Planning Planning Testing
LCG RENKEI
KEKCC supports both LCG and NAREGI/RENKEI
Many Japanese HEP groups are joining LCG
NAREGI middleware is being deployed as the general purpose e-science infrastructure in Japan
RENKEI is developing a system to provide a seamless user environmentbetween the local resources and multiple grid environment
GRID for ILCTwo Vos have been used:
CALICE-VO:Test beam data analysis and MC. Standard data processing in GRID
ILC-VO:Needs huge CPU resources for the studies. Available only on GRIDStandard MC samples ( ~ 50TB) are on GRID for sharing
Status:A typical data transfer rate from IN2P3/DESY to KEK: ~ 200kB/sec/port
a frequent timeout for transfer of ~ 2GB: Cured by removing a time out at IN2P3
Overhead of catalog accessILD DST: many small size DSTs, limited by CPU time for a MC job.MC and DST production at DESY/IN2P3
Merge DSTs to create a large size file, then replicated to KEK
A typical GRID performanceFile transfer: IN2P3 Kobe, 184 files/210 GB in 13 hours
- part of ILD LOI study, in Dec. 2008- 10 ports/job
Pedestal in transfer time ~ 20~60sec. < 100MB is not effective.Instantaneous transfer rate: average 4 MB/sec, Max. 10 MB/sec
not great, but has been used for real works successfully
Data size vs Time Transfer rate
During Dec. ‘08 to Feb. ’09, O(10TB) data have been exchanged through GRID.It was crucial for the successful LOI studies.During Dec. ‘08 to Feb. ’09, O(10TB) data have been exchanged through GRID.It was crucial for the successful LOI studies.
ILD
DESY Computing Seminar
Recent issues on GRIDAfter LOI, KEK has extended GRID resources
CPU: 0.3M SI2K to 6M SI2K ( utilize old CPU resources )Storage: IBM HPSS as the backend storage. Tape capacity up to 3 PB.shared by batch server and many other groups.
Operational issuesnetwork speed outside JapanWith increasing WNs, many new problems seems to appear
Files in HPSS : file transfer breaks frequentlyAccess to MySQL server from WNDisk space of WNs are not sufficient for storing large temporary dataOver-loaded WMSVery slow turn around from job submission to job out retrievingFailure rate is not low enough and system tuning is yet to be done.
4312 July 2007 Akiya Miyamoto
ILD
DESY Computing Seminar
SummaryVery extensive developments and studies have been done using ILCsoftware tools.We are aiming to produce DBD by 2012 and software based study will play crucial role in it.Our main efforts right now are
updates of simulator modelsimprovements of core software tools
Improvements of reconstruction tools and new benchmark studies will follow soon
4412 July 2007 Akiya Miyamoto