simulation and pfa development efforts at niu: status & plans d. chakraborty, for the niu ilc...

Simulation and PFA Simulation and PFA Development Efforts at NIU: Development Efforts at NIU:

Status & PlansStatus & PlansD. Chakraborty, for the NIU ILC detector

group

Main contributors:

V. Zutshi, G. Lima, J. McCormick, R. McIntosh

Fermilab, 22 Sep 2005

Fermilab, 22-09-05 Simulation & PFA at NIU Dhiman Chakraborty

2

OutlineOutline• Simulation

– GEANT4-based packages: LCDG4, TBMokka, (SLIC)

– Parametric simulation of GeV-to-ADC: DigiSim

• Particle-Flow Algorithms

– Density-weighted Clustering in Calorimeter

• Calorimeter-only (no track-seeding)

• Same for ECal (e, , and HCal (h+, h0 ).

– Replace cal clusters with matching (MC) tracks, if any.

• The Directed Tree Algorithm

– Association of isolated “fragment”s or “satellite”s.

• Work in Progress and Future Plans


3

Simulation: SummarySimulation: Summary• Primarily interested in exploring the (semi-)digital

hadron calorimeter option in general, with scintillator as the active material in particular.

• When we started, there was no GEANT4-based simulation package to produce output compatible with the Java-based reconstruction framework.

• LCDG4 was written at NIU starting from a LCDROOT developed at U. Colorado. LCDG4 fully conformed with all standard I/O requirements (geometry input & event output formats), offered new features like non-projective geometry & better links to MC truth, fixed many bugs. (GL,RM)


4

Simulation: Summary Simulation: Summary (contd.)(contd.)• NIU collaborated with DESY,LLR to produce

TBMokka, derived from Mokka (the G4-based TESLA simulator) for simulation of the CALICE TB module. NIU also produced a stand-alone G4-based application for cross-check. (JM)

• Thoughts on what is now SLIC started while JM was at NIU. JM was on SLAC-NIU joint appointment when SLIC was written.

• DigiSim allows easy parametric simulation of the many effects between the G4 energy deposits and recorded data. Available in Java & C++. (GL)


5

TasksTasks

• Full-detector simulation:

LCDG4 (for ALCPG)

• Test-beam module simulation:

TBMokka (with DESY/LLR, for CALICE)

• Simulation of detector imperfections:

DigiSim (for ALCPG, CALICE, …)


6

Full detector simulation:Full detector simulation: LCDG4LCDG4

• GEANT4 v7.0.• Most hadronic physics lists available,

LCPhysics not tested yet.• Input format: binary STDHEP• Output format: .lcio (standard) or .sio

(Gismo compatible)• Nominal American detector geometries

are implemented via XML geometry files (continuous tubes+disks only).


7

LCDG4: some nice LCDG4: some nice featuresfeatures

• Correct MC particle hierarchy, even when V0s and hyperon decays are forced at the event generation stage.

• Energies deposited in absorbers are available for cross-checks.

• Non-projective geometries available.

• Simple analysis code and documentation available from CVS and http://nicadd.niu.edu/~lima/lcdg4/.


8

LCDG4: Projective vs. non-LCDG4: Projective vs. non-projectiveprojective

• Barrel only.

• Rectangular virtual cells with linear dimensions, controlled at run time (XML).

• Ecal and HCal can be made projective or non-projective independently.

projective non-projective

9

LCDG4: XML geometry descriptionLCDG4: XML geometry description

Example: barrel HCal (dimensions in cm).

...<volume id="HAD_BARREL" rad_len_cm="1.133" inter_len_cm="0.1193"> <tube> <barrel_dimensions inner_r="144.0" outer_z="286.0" /> <layering n="34"> <slice material="Stainless_steel" width="2.0" /> <slice material="Polystyrene" width="1.0" sensitive="yes“ /> </layering>  <cell_info length="1.0" height="1.0" offset="0." outer_r="246.5"/> </tube> <calorimeter type="had" /></volume>...

Most of detector changes are very easy to make!

Proj or NP selectioneither one, not both.


10

LCDG4 output: general LCDG4 output: general featuresfeatures

• One particle collection and several hit collections (one hit collection per subdetector)

• Each hit points to the contributing particles (except tracker hits from calorimeter back-scatters).– LCIO only: (x,y,z) coordinates stored for every hit

• All secondaries above an energy threshold (now set at 1 MeV), except for shower secondaries, are saved in output with:– Particle id and generation/simulation status codes.– Production momentum, production* and ending position

(*LCIO only).– Calorimeter entrance: position and momentum (SIO

only).– Pointers to parent particles.


11

LCDG4: LCDG4: zoom in on primary zoom in on primary interactioninteraction

0

0 →0 0

0

ΛK

s0

Decaysforced ingenerator correctlyprocessed


12

LCDG4: example: LCDG4: example: e+e- tt


13

LCDG4: wrap-upLCDG4: wrap-up• Being replaced by SLIC as the the

ALCPG standard, but many analysis packages have yet to adapt to the new standards.

• Code with doc + instructions available:http://nicadd.niu.edu/~lima/lcdg4

• Many interesting single particle and full event samples available, new ones can be requested:

For detailed access instructions, see http:// nicadd.niu.edu/~jeremy/admin/scp/index.html

http://nicadd.niu.edu/digisim/DigiSim.html


14

DigiSimDigiSim• Goal: a program to simulate the signal collection and

digitization process in the ILC detector(s).

• Converts the GEANT4 output (energy depositions and space-time coordinates) into the same format AND as close as possible to real data from readout channels.

• Same code can then be used to process MC & real data.

• Serves a convenient and standardized hook for the user to model such detector effects as inefficiencies, non-uniformities, non-linearities, attenuation, noise, cross-talk, hot and dead channels, cell ganging, etc. involved in signal collection, propagation, and conversion/recording.


15

Requirements and Requirements and choiceschoices

• Basic requirements:– Object-oriented design to simplify

maintenance and implementation of new functionality

– Should be usable for both ALCPG’s Java-based reco framework and CALICE’s MARLIN (C++)

– To be tested for TB simulation– Usable in stand-alone and preprocessor modes

• In stand-alone mode, it produces LCIO events in the sae format as real data. All input information is preserved.

• In preprocessor mode, doesn’t produce any persistent output, event in memory passed on to reconstruction.

– Very easy to configure and enhance through text-based driver files.


16

DigiSim event loopDigiSim event loop

Modifiers

TempCalHitsSimCalo.Hits TempCalHits RawCalo.Hits

DigiSimProcessor

SimHitsLCCollection RawHitsLCCollection

● Calorimeter hits shown here, but applies just as well to other subdetectors.

● TempCalHits are both input and output to each modifier

● All processing is controlled by a DigiSimProcessor (one per subdetector)

● Modifiers are configured at run time, via the Marlin steering file

● New modifiers can be easily created for new functionality.


17

Example modifiersExample modifiers

threshold

Smearedlinear

transformation

GainDiscrimination is a smearedlin.transf. + threshold on energy

SmearedLinear is a func-basedsmeared linear transformationon energy and/or timing

Einput

Eoutput


18

DigiSim example: Scint. HCalDigiSim example: Scint. HCal1. Photodetection1. Photodetection


19

DigiSim example: Scint. HCalDigiSim example: Scint. HCal2. Noise and discrimination2. Noise and discrimination

Mip peak ~ 15 PEs


20

Scintillation: 104 / MeV

Photodetector QE: 15%

Effcoll

= 0.0111 +/- 0.0020

Expo + gauss noise

¼ MIP threshold

HCal scintillator digitization (prelim.)HCal scintillator digitization (prelim.)

simulateddigitized


21

Scintillation: 104 / MeV

Photodetector QE: 15%

Effcoll

= 0.0111 +/- 0.0020

Expo + gauss noise

¼ MIP threshold

HCal scintillator digitization HCal scintillator digitization (prelim.)(prelim.)


22

DigiSim: Usage DigiSim: Usage InstructionsInstructions• Download/install/build java 1.5, Maven, org.lcsim

(see http://www.lcsim.org for details)

• Drivers needed: (all available from org.lcsim.digisim)CalHitMapDriver, DigiSimDriver and CalorimeterHitsDriver, plus LCIODriver (for standalone run, saving output file) or YourAnalysisDrivers (as an on-the-fly preprocessor)

• DigiSim configuration file stored on LCDetectors: digisim/digisim.steer

• Run it:– From command line: after setting the CLASSPATH (see

docs for details)java org.lcsim.digisim.DigiSimMain <inputfile>an output file ./digisim.slcio will be produced, to be used for analysis or reconstruction

– From inside JAS: DigiSimExample is available from Examples -> org.lcsim examples

http://www.lcsim.org/


23

Particle Flow Algorithm Particle Flow Algorithm DevelopmentDevelopment

• ECal:– 30 layers, silicon-tungsten.

– 5mm x 5mm transverse segmentation.

• HCal:– 34 layers, scintillator-steel

– 1 cm x 1 cm transverse segmentation.

• Magnetic field: 5T

• Support structures, cracks, noise, x-talk, attenuation, inefficiencies,… not modelled.


24

++ energy resolution as function energy resolution as function of energy for different (linear) of energy for different (linear)

cell sizescell sizes


25

++ energy resolution vs. energy resolution vs. energyenergy

Two-bit (“semi-digital”) rendition offers better resolution than analog


26

Nhit vs. fraction of Nhit vs. fraction of ++ E in cells with E in cells with

E>10 MIP:E>10 MIP: Gas vs. scintillatorGas vs. scintillator

2-bit readout affords significant resolution improvement over 1-bit for scintillator, but not for gas


27

++ energy resolution vs. energy resolution vs. energyenergy

Multiple thresholds not used


28

Non-linearityNon-linearity

• In scint, Nhit/GeV varies with energy.

• This will introduce additional pressure on the “constant” term.

• For scintillator, the non-linearity can be effectively removed by “semi-digital” treatment.


29

±± angular width: density angular width: density weightedweighted


30

Simulation SummarySimulation Summary• Large parameter space in the nbit-

segmentation-medium plane for hadron calorimetry. Optimization through cost-benefit analysis?

• Scintillator and Gas-based ‘digital’ HCals behave differently.

• Need to simulate detector effects (noise, x-talk, non-linearities, etc.)

• Need verification in test-beam data.

• More studies underway.


31

Clustering Clustering (~2 yrs now)

• Seeds: maxima in local density:

di = (1/Rij)

• Membership of each cell in the seed clusters decided with a distance function.

• Only unique membership considered.

• Calculate centroids.

• Iterate till stable within tolerance.


32

Separability of clustersSeparability of clusters

Best separability is achieved when width of a cluster is small compared to distances between clusters.

J = Tr{Sw-1 Sm}


33

Separability of clusters Separability of clusters (contd.)(contd.)

where

Sw = Si WiSi

Si = covariance matrix for cluster ci (in x, y,

z)Wi = weight of ci (choose your scheme)

Sm = covariance matrix w.r.t. global mean


34

Two (parallel) Two (parallel) ++’s in TB sim:’s in TB sim:

cmcm

cmGeV


35

Two (parallel) Two (parallel) ++’s in TB prototype sim:’s in TB prototype sim:separability (separability (JJ) vs. track distance) vs. track distance

for different cell sizesfor different cell sizes


36

DHCal: Particle-flow algorithm DHCal: Particle-flow algorithm (NIU)(NIU)

• Nominal SD geometry.

• Density-weighted clustering.

• Track momentum for charged,

• Calorimeter E for neutral particles.

+ p0

Cal only

PFA


37

DHCal: Particle-flow algorithm DHCal: Particle-flow algorithm (NIU) (NIU)

Photon Reconstruction inside jets Photon Reconstruction inside jetsExcellent agreement with Monte Carlo truth:


38

PFA Jet Reconstruction PFA Jet Reconstruction summary summary (old)

• Cone clustering in the calorimeters,• Flexible definition of weight (energy- or

density-based),• Generalizable to form “proto-cluster”

inputs for higher-level algorithms. • Replace cal clusters with matching MC

track, if any.• Based on projective geometry.• New clustering algorithms available.


39

DHCal: Particle-flow algorithm DHCal: Particle-flow algorithm (NIU)(NIU)

Reconstructed jet resolutionReconstructed jet resolution

Cal only Digital PFA

ZZ 4j events


40

The Directed Tree The Directed Tree AlgorithmAlgorithm

• Define neighborhood for a cell

• Discard cells below threshold (0.25 MIP)

• Calculate density for each cell i

• If(density==0) ?

else

calculate (Dj – Di )/dij

where j is in the neighborhood

find max []


41

The Directed Tree Algorithm (contd.)The Directed Tree Algorithm (contd.)

• If max[] is –ve

i starts a new cluster

if max[] is +ve

j is the parent of I

if max[] == 0

avoid circular loop

attach to nearest


42

Single hadrons in the ECalSingle hadrons in the ECalGenerated clusters

Reconstructed clusters

Clusters from single hadrons are reconstructed well.Some “fragment”s or “satellite”s remain unassociated.


43

EM clusters in ZEM clusters in Zqq Eventsqq EventsGenerated clusters

Reconstructed clusters

Only a few highest-E clusters shown.


44

The confusion termThe confusion term• Internal to calorimeter.• Reconstruct “gen” and “rec” clusters,• A “gen” cluster is a collection of cells which

are attached to a particular MCparticle. All detector effects are included in this cluster.

• Find centroids and match to nearest “rec” cluster, making sure that no cluster gets associated twice.

• Somewhat conservative.


45

ZZqq Eventsqq Events

• Calculate Erec/Egen for each generated cluster

• Enter into histogram with weight Egen/Etotal.


46

Cluster Matching and MergingCluster Matching and Merging

• Stage 1: one-to-one gen-reco matching

based on distances (3D or angular)

→ unassociated clusters (“satellites”)

• Stage 2: attach satellites to reco clusters

based on angular distances: possible cuts

on angular separation, satellite energies,

number of hits,...


47

Preliminary ECal AnalysisPreliminary ECal Analysis• 500 events, with 2-

pions 10 cm apart at ECal face, using SDNPHOct04 detector

• neighborhood definition: (dφ=5, dz=5, dlayer=9)

• discard events with decays or interactions before Ecal

• Look at:– eratio1: Erec/Egen after

stage 1 (matching)

– eratio2: Erec/Egen after stage 2 (merge satellites)


48

Preliminary HCal AnalysisPreliminary HCal Analysis• 500 events, with 2-pions

10cm apart at Ecal face, using SDNPHOct04 detector

• neighborhood definition: (dφ=2, dz=2, dlayer=2)

• discard events with decays or interactions before Ecal

• Look at:– eratio1: Erec/Egen after

stage 1 (matching)

– eratio2: Erec/Egen after stage 2 (merge satellites)


49

Current StatusCurrent Status• Analysis of complex events shows some

problems with too many satellites – how to

associate them with the right parent shower?

• Clustering algorithm ported to org.lcsim, to be

certified. Committed to LCSim CVS repository.

• More manpower for the PFA development effort.

• Work in progress, a lot to do ...


50

Current Status (contd.)Current Status (contd.)

With Perfect PFA (no confusion term), σ(E) = 3.1 GeV

Zqq

σ(E) = 4.3 GeV

•Scint, 1 cm x 1cm. •Density-weighted clustering, but analog E calculation in both ECal & HCal. •Min 5 cells for cluster. •Cell threshold = 0.5 MIP. •Satellite attachment not used.•No cut on theta.


51

PlansPlans• No plans to get involved in any new simulation

development work...• Will continue to support LCDG4, but no further development• Will further develop DigiSim – a long-term product.

• ... until, perhaps, we are ready to parametrize some of the simulation (see below)

• Will concentrate more on algorithm development (with DigiSim). • New ideas on reducing the confusion term and accounting

for differences in low-E neutral hadron energy sampling need to be coded, polished.

• Enormous volumes of MC will have to be generated and analyzed to get a handle on these.


52

Plans (contd.)Plans (contd.)• Major re-writing of code underway to make it

more object-oriented.

• Aim to be ready to analyze the TB data as soon as they become available.

• A great deal of work to do.

• Coordinated, efficient use of scarce resources is key to success.


53

Backup slidesBackup slides

simulation and pfa development efforts at niu: status & plans d. chakraborty, for the niu ilc...

Documents

simulation pfa

detector simulation

niu dhiman chakraborty

easy parametric simulation

gl slide

rm slide

nonprojective barrel

mcintosh fermilab