Iacopo Vivarelli On behalf of ‡

Post on 15-Jan-2016

49 views 0 download

description

Jet Energy Correction With Cell Energy Density and Layer Weighting † The Way from Rome -> Costa Brava -> Milano -> Now -> First Data. Iacopo Vivarelli On behalf of ‡ T. Costin, A. Gupta, P. Francavilla, F. Merritt, M. Oreglia, F. Paige, J. Proudfoot, C. Roda, B. Salvachua, I.Vivarelli. - PowerPoint PPT Presentation

Transcript of Iacopo Vivarelli On behalf of ‡

1

Jet Energy Correction WithCell Energy Density and Layer Weighting†

The Way from Rome -> Costa Brava -> Milano -> Now -> First Data

Iacopo VivarelliOn behalf of‡

T. Costin, A. Gupta, P. Francavilla, F. Merritt, M. Oreglia, F. Paige, J. Proudfoot, C. Roda, B. Salvachua, I.Vivarelli

† JetCalib: the package used to determine weights used in the so-called Global (H1-style) Jet Calibration‡ Thanks to Peter Loch, Rolf Seuster, Peter Van-Gemmeren, Hong Maand Scott Snyder for help with software issues.

2

Outline

• Reminder – Factoring Jet Energy Correction• Basis for comparison (Release 12) vs today

(Release 14, GEANT Scale changed)• Performance on 12 and 14 and some validation

checks• Software layout• Work in Progress

– Integrate cell density and layer weighting– Layer weights and layer weighting on AOD– Scale using inversion technique

• Conclusions

3

Factoring Jet Energy Correction – H1-style

Step 1:

Correction based on cell energy density - expected to be mainly sensitive to e/h (but not entirely separated).

Correction for dead materials (material upstream of calorimeter, cryostat, tile gap regions)

- These corrections are derived from chi-square min.

for jets in good calorimeter region.

Step2:

Jet Et and eta dependent scale correction for cracks and

gaps in the calorimeter

Step3:

Additional physics related corrections for e.g correction

based on fraction of energy in LAr calorimeter (Fem) will be sensitivity to MC hadronization model.

CellWt( E/V, Layers )

MatWt(PreSamp, EM1)CryoWt( EM3,Tile1)GapCalWt( TileGaps )JetScale( JetEt , JetEta )

JetEnergy( fem )JetEnergy(Ftrk)…

4

Multi-jet event - Cone 0.7 - Use towers that include only those cells that belong to topoclusters - H1 weights.

(E)

E

0.94

E(GeV ) 0.05

0.0

E(GeV )

(E)

E

0.72

E(GeV ) 0.02

2.3

E(GeV )

Un calibrated (barrel region)

Calibrated (towers from topos)

)(

9.202.0

)(

74.0

E

)(

GeVEGeVE

E

Calibrated (standard towers)

From: http://indico.cern.ch/conferenceDisplay.py?confId=a062339

12x: H1-Style & QGSP (Reminder from Costa Brava)

Our reference to assess improvements in the method and software

5

Athena 14x (H1-style equivalent to12x.)

http://indico.cern.ch/conferenceDisplay.py?confId=41369

With new weights, linearity

shows similar performance to that obtained in

Rel 12.

QGSP_BERT

-> 5% Scale shift wrt EMV

New Weights now available by JobOption

6

Test Using Independent Physics Sample

• TTbar MC@NLO Herwig vs Pythia di-jet: different in physics process, generator and fragmentation

See response deviation from unity at the 1-4% level when applying weights to Cone7 and Cone4topo jets: is this physics or calibration?

Note: Test done using our full chain: drop new weights into database, run full reconstruction chain, produce plots via JetPerformance Pkg

+ 2%

- 2%

7

Testbeam vs Testbeam MC: Performance on energy linearity

MC (QGSP_EMV) and Data are within 4-5% @ HAD Scale

Redoing the exercise for QGSP_BERT

References:Atlas SM meeting 22 March 2007Hadronic Calibration workshop Milano, 26-27 April 2008CSC Note - Jet, Missing Et and Tau Combined Performance: Detector Level Jet Corrections

•H1 Weights derived on MC simulation of single pions with the CTB04 geometry and applied both to MC and real data.

•Obtain linearity and resolution.

•Compare performance between MC and real data.

8

JetCalib Software and JetSampling from Athena 12.x to 14x

• In release 13– EM, H1, Pisa, Sampling integrated into a common

framework– Level 2 calibration based on JetSampling approach

• Release 14 development– Changes to accommodate ESD, AOD, DPD, Truth,

Reco in common framework– Improved algorithms to determine weights and global

scale– Performance

9

ESD, AODCalibrated JetsCells, Layers

Truth Jets

Jet SamplingCollection

ASCIICalibrationconstants

CalibratedJet Sampling

Collection

BuildJetSamplingAlg

JetFit

H1Calibrator,LayerCalibratorAlg

CBNTAA_JetSampling

CalibrationConstants in

DataBaseJetAlgorithm

BuildJetSamplingTool

JetCellCalibratorToolJetLayerCalibratorTool

JetClassifier

ROOT Jet Sampling

ntuple

ROOT macros

JetPerformance

ESD, AODCalibrated JetsCells, Layers

Truth Jets

Software Layout used for H1-style/Sampling Calibration

New developments

1. Exploiting the best from combining ideas from Cell and Layer weighting strategies

2. Using Layer calibration strategy from AOD

3. Inversion technique to improve performance at low pt

10

11

Combining H1-Style and (Layer) Sampling MethodsNew development 1

H1-style uses cell E/V to discriminate EM

and HAD showers. Weights are independent of jet energy.

Sampling method uses the longitudinal shower profile. Weights are a function of jet energy and eta.

These ideas can be combined

- Regroup the cells in finer longitudinal

layers.

-- Also avoid mixing CaloCells of different physical/readout sizes.

- Use fem bins etc., at JetScale stage or later.

12

Athena 14x (combining H1 & Sampling)

http://indico.cern.ch/conferenceDisplay.py?confId=41369

Better optimization of H1 Fit Regions•Separate EM2 from EM3 (EMB is already separated in two layer at ||=0.8. EME at ||=2.5)•Fit HEC(0,1) and HEC(2,3) in separate E/V bins (HEC layers divided in two layers at ||=2.5, since the readout size changes).•Fit Tile0 and Tile1 in E/V bins separately. •Fit Tile2 as a single layer.

First attempts:

Linearity looks good

Some improvement in resolution

Further improvement possible at JetScale stage:

Better use of Gap scintillators

Apply CellWt to cells above a noise threshold

13

Derive and Apply Layer Weights at AOD level- Release 14.2.20

Cone4Tower Jets

• Resolution:

E

cb

E

a

E

(GeV)

Code works – now looking at performance

Calorimeter longitudinal layers combined together depending on the jet pseudorapidity

Jets classification based on calorimeter em fraction and jet pseudorapidity

Layer weights depend on the jet energy. Determined minimizing the resolution

14

Current status of layer correction: Linearity

Cone4Tower2008 AOD

Cone4Tower2006 ESD

ATL-COM-PHYS-2006-062

Work in Progress

Under Study

15

Current status of layer correction: Resolution

0.04

Cone4Tower2008 AOD

Cone4Tower2006 ESD

ATL-COM-PHYS-2006-062

Strange poor resolution improvement on || (1.5,2.5) under study

Work in Progress

16

Revisit an old idea: Truth->Reco inversion technique

New development 3(http://indico.cern.ch/getFile.py/access?contribId=s1t2&resId=0&materialId=0&confId=a057453)

also being used in more recent work, for example http://indico.cern.ch/getFile.py/access?contribId=5&resId=0&materialId=slides&confId=45187

Use following conditional probability to estimate corrected jet energy.P(ET | ER) = P(ER | ET )*P( ET )/(Normalization)

ER is reconstructed jet energy. ET is true jet energy. P( ET ) is input jet spectrum.

We know P(ER | ET ) from MC - This is the response of the detector to a truth jet of certain energy and eta.

Want to Determine Jet Scale based on Reconstructed Jet Energy

Dips in the sample are associated with Jx thresholds

Weighting by cross section improves linearity but not completely and introduces other issues

17

Inversion Technique: Apply to EM Scale Jetshttp://indico.cern.ch/getFile.py/access?

contribId=20&sessionId=5&resId=0&materialId=slides&confId=41483

Response function built

from em-scale Jets.

- Truth jet pT >20 GeV.

- Only five eta bins 0.0- 0.7,0.7-1.5,1.5-2.5. . .

Applied to E_reco with

matched truth pT >30 GeV.

Ideally we would want to build response function in finer bins of eta and

energy.

As noted by others, the resolution is also

improved

0< ||<0.7

18

Apply Also to H1-Style Calibrated JetsHere the response functi-on is built from jet applied with CellWt(E/V).

Strong improvement in re- solution compared to em-scale jets (last slide).

Note: For comparison the

constant terms are fixed

in the lower fits to upper

ones.

0< ||<0.7

19

Conclusions…• QGSP_BERT simulation agrees better with test beam. Gives O(5%) shift in

jet scale compared to QGSP_EMV.

• Is simulation converging? Then worthwhile to redo jet energy scale.

• Define jet scale by comparison with same jet algorithm(s) on MC truth. (Need additional corrections for physics results).

• First simply repeated H1 and layer weights fits with QGSP_BERT.

• Agreement for pythia QCD jets and MCAtNLO top samples comparable to earlier results.

• But higher statistics few percent effects clearly visible. Still workong to resolve these.

…and next steps

• Work in progress– Integrate cell density and layer weighting– Layer weights and weighting on AOD – Scale using inversion techniques

• Crucial to develop in situ techniques to verify/improve jet simulation and response

• Should extend work to new jet algorithms

• Software now more integrated and better organized. It should be more usable by others. Still need more documentations

21

Backup

22

Performance on energy linearity

Hadronic scale

EM scale

MC

Is the performance on linearity consistent between MC and data ?

23

Next steps

Redo this analysis using the latest data analysis and MC with QGSP_BERT

Use this method to evaluate the performances of other calibration strategies: Sampling, LC …

Understand the possibility of using E/p in ATLAS to do the same analysis.

24

Software location: Reconstruction/Jet

• JetEvent: Definition of JetSampling and JetSampling Collection. JetSampling class contains the energy in layers, energy density in cells, radial profiles, truth jet (PIC and NTJ), uncalibrated and calibrated jet. All this information is needed by JetFit algorithm to calculate the calibration constants

• JetEventAthenaPool: Transient/persistent converters of JetSamplingCollection, JetSamplingCollectionCnv

• JetEventTPCnv: Persistent representation of JetSampling and JetSamplingCollection

– JetSampling_p1 and JetSamplingCnv_p1 class– JetSamplingCollection_p1 and JetSamplingCollectionCnv_p1

• JetUtils: Definition of JetClassifier class.JetClassifier implements all methods to fill JetSampling.

• JetRecTools: Contains the tools to calibrate the jets. – JetSamplingCalibTool: Calibrateds jets using Layer method, it will change name to

JetLayerCalibratorTool– JetCellCalibratorTool: Calibrates jets using H1-style method

JetRecTools used to hold the algorithm to create JetSampling from ESD (JetSamplingCalibAlg) now this algorithm has been moved to JetCalib with the name BuildJetSamplingAlg

25

Software location: Reconstruction/Jet

• JetCalib: Main package for the calibration.– BuildJetSamplingTool and BuildJetSamplingAlg: Algorithm and Tool

used to create the JetSampling collection from any pool file.– JetFit: Athena algorithm that implements the calculation of the

calibration weights. Several configurations are foreseen.• Different eta ranges• Different energy binning• Different calibration methods, the most used:

– FitSampleIn2E3FemBin: Layer or Sampling method ( atl-com-phys-2006-062 )– FitCellDenH1Style: H1-style method (cell energy density)– FitCellDenInLayers: new H1-layer style method (cell energy densisty in layers)

– H1Calibrator and LayerCalibratorAlg: Athena algorithms that read the calibration from ascii files and apply it to a JetSampling collection.

– CBNTAA_JetSampling + macros: Athena algorithm to create a ROOT ntuple from a JetSampling collection. In the share directory there are several macros to produce linearity and resolution plots.

• JetPerformance: Used to produce histograms used for CSC note and test the performance of the calibration

26

JetSampling Collection: ContentsSummary of all the variables accessible from JetSampling:

– Kinematics of the reconstructed jet at EM scale– Kinematics of the Nearest-truth-jet– Kinematics of the Particle-in-cone jet– Kinematics of the calibrated jets:

• H1• PISA• SAMPLING

– Distantance of the 1st and 2nd Nearest-truth-jet– Jet layer information:

• PreSamplerB, PreSamplerE• EMB1, EMB2, EMB3• EME1, EME2, EME3• TileBar0, TileBar1, TileBar2• TileExt0, TileExt1, TileExt2• TileGap1, TileGap2, TileGap3• HEC0, HEC1, HEC2, HEC3• FCAL0, FCAL1, FCAL2

– Jet energy in the EM calorimeters– Jet energy in the HAD calorimeters– JetSums– JetECS (still needs some cleanup0– Energy in cone radii: Radial profiles

27

JetSampling from ESD, AOD, DPD• ESD:

– Reconstructed jet– Layers, Radial profiles, Energy density– Truth information

• AOD:– Reconstructed jet– Layers– Nearest-truth-jet– Particle-In-cone (from GEN_AOD instead of TruthEvent?)

• DPD ( from data ):– Reconstructed jet– Layers– Radial profile– Energy density– … Same as ESD except for the truth

28

Job Options for use in Applying Weights to EM Scale jets

29

Combined Test Beam 2004 data may be used to investigate the performance of any hadron calibration on real single hadrons.

Method:

H1 Weights derived on MC simulation of single pions with the CTB04 geometry and applied both to MC and real data.

Obtained (for example) the linearity and the resolution.

Compare performance between MC and real data.

Work done up to now:

Used MC QGSP-GN

Hadron Calibration: Standard H1 cell calibration

Single Pions at η=0.35, 20 → 350 GeV;

Quality cuts on the beam position;

Cuts to reject muon and electron contaminations;

Estimate of the effect of the proton contamination (for π+ ).

Monte Carlo Simulation: Comparison to Testbeam

30

Layer weighting on AODNew Development 2.

• Region selection:– ||(0.0, 1.5)

Layer 0 = PreSampler + LArEM1Layer 1 = LArEM2Layer 2 = EM2 + Tile1Layer 3 = Tile2 + Tile3

– ||(1.5, 3.2)Layer 0 = PreSampler + EM1 + EM2Layer 1 = EM3 + HCAL + FCAL

– ||(3.2,4.4)Layer 0 = Full jet energy

• Jets classified in bin of , energy and fractional energy (fem)

• Dependence on the energy as:

• Minimization function:

layers

1

Recjets

1

RecRef2RecRef ,n

iiin

m

jnnnn EwEEEEES

EnergyJet

EM2EM1PreSampler fem

cutE

Ebaw log