New Cell Energy Density weights and scale factors parameterization MC09

Post on 02-Jan-2016

15 views 0 download

Tags:

description

New Cell Energy Density weights and scale factors parameterization MC09. Sebastian Eckweiler Institut fur Physik Johannes-Gutenberg-Universitaet. Belen Salvachua High Energy Physics Division Argonne National Laboratory. The Goal. Calculate new Global Cell Energy Weights for: - PowerPoint PPT Presentation

Transcript of New Cell Energy Density weights and scale factors parameterization MC09

New Cell Energy Density weights and scale factors parameterization

MC09

Belen Salvachua

High Energy Physics DivisionArgonne National Laboratory

Sebastian Eckweiler

Institut fur Physik Johannes-Gutenberg-Universitaet

2

The Goal

Calculate new Global Cell Energy Weights for:

– New MC09 simulation

– Test weights for AntiKt jets and Cone jets Calculate new Re-scale factors with Num. Inv. technique for all jet

collections on new AODs Store constants into database and use them as default for MC09

3

Use Validation Samples to calculate the weights

Dataset Name Events Available

mc09_valid.105009.J0_pythia_jetjet.recon.ESD.e344_s561_r731/ 91000

mc09_valid.105010.J1_pythia_jetjet.recon.ESD.e344_s561_r731/ 64250

mc09_valid.105011.J2_pythia_jetjet.recon.ESD.e344_s561_r731/ 61000

mc09_valid.105012.J3_pythia_jetjet.recon.ESD.e344_s561_r731/ 98250

mc09_valid.105013.J4_pythia_jetjet.recon.ESD.e344_s561_r731/ 98250

mc09_valid.105014.J5_pythia_jetjet.recon.ESD.e344_s561_r731/ 99000

mc09_valid.105015.J6_pythia_jetjet.recon.ESD.e344_s561_r731/ 100000

mc09_valid.105016.J7_pythia_jetjet.recon.ESD.e344_s561_r731/ 97500

mc09_valid.105017.J8_pythia_jetjet.recon.ESD.e344_s561_r731/ no events available

Thanks to Iacopo Vivarelli for the production of these samples

4

PROBLEM: Retrieving EM scale cell energy

AntiKt jets were not in the validation ESDs

– need custom reconstruction Use signal state to retrieve EM scale:

– Tower jets seem fine

– Topo jets were providing incorrect Cell EM energy

AntiKt4Topo AntiKt4Tower

Extra weight for Topo jets energy

5

Problem and Solution

Problem:

– By default make_StandardJetGetter('AntiKt',0.4,'H1Topo').jetAlgorithmHandle() was taken Calibrated Topo Clusters

Solution : Specify input collection Uncalibrated Topo Clusters

– myalg0=make_StandardJetGetter('AntiKt',0.4,'H1Topo').jetAlgorithmHandle()

– myalg0.InputCollectionNames=['CaloTopoCluster']

And commented out: doTopoClusterLocalCalibration = False Jobs were re-sent to the grid Correct EM energies !!!

Thanks to Pier-Olivier and Pierre-Antoine

6

New H1 weights !

---- AntiKt4Topo

---- AntiKt4Tower

---- Cone4Topo

---- Cone7Tower

Only large differences for AntiKt4Tower in very low statistic region

Previous releases we used Cone7Tower weights for ALL collections.

Since we are mainly using now AntiKt, We propose to keep AntiKt4Topo weights for ALL MC09 collections.

7

Linearity and resolution without Num. Inversion

AntiKt4Topo

NO num. inversion

8

Numerical Inversion

We decide to use AntiKt4Topo H1 constants for all collections However, the Numerical Inversion fits provided by Sebastian show

difficulties to find the proper parameterization.

– Large fluctuations on the function. That did NOT happen before (see next 2 slides)

9

OLD H1 (from Cone7Tower): AntiKt4Topo

10

New AntiKt4Topo H1

11

Problem of fluctuations with new constants OLD constants calculated with Cone7Tower and applied to all Jet Collections NEW constants calculated with AntiKt4Topo Difference could be due to:

– Cone size

– Topo vs Tower

– Change on the simulation/reconstruction We re-did the fits for new constants for Cone7Tower

– Similar to OLD MC08 Cone7Tower NOT due to simulation/reconstruction

New H1

From AntiKt4Topo

New H1

From Cone7Tower

12

New H1

From AntiKt4Topo

New H1

From Cone7Tower

New H1

From AntiKt6Topo

New H1

From AntiKt6Tower

Change Cone size and input

Difference due to:

Input constituents

13

Between AntiKt 4 or AntiKt 6:

Differences are minimal.

AntiKt4Topo weights for MC09

14

Numerical inversion in mc09

Applying the new mc09 cell weights: Improved response:

– Quicker rise and closer to 1

– But: significant changes in shape• increased data - fit discrepancy

Several disadvantages inprocedure with fitted histograms:

– no physical motivation

– Shape depends on η-ranges:• would need several

parameterizations• polynomial up to arbitrarily high

orderalso unsatisfactory

• difficult to automatize

15

Results - Mc09

Discrepancies directly propagated to response Nevertheless, scale within ± 2% :

– If physical fluctuations larger deviations in different sample expected

16

Looking for improvements

Basicall there are 2 classes of possibilities:

– Stick to parameterizations:- needs lots of manual effort needed+ only trivial changes in software needed

– Direct use of TObjects / TGraphs:(Get scale factor with h->GetBinContent() or similar)

+ would be easier to automate- eventually sensitive to statistical fluctuations

Possible combination of both methods: smoothed Graphs

17

Looking for improvements

TGraphSmooth offers several methods to smooth Graphs Easiest possibility: smooth points using a given Kernel-function p(x,y)

– Smoothed values:

Example for p(x, xi):Gaussian with mean x

Kernel-function parameters needsome adjustment:e.g. width of the Gaussian

Sensible ‚upgrade‘:p(x,xi) -> p(x,xi) / σito incorporate uncertainties

18

Looking for improvements

TGraphSmooth offers several methods to smooth Graphs Easiest possibility: smooth points using a given Kernel-function p(x,y)

– Smoothed values:

Example for p(x, xi):Gaussian with mean x

Kernel-function parameters needsome adjustment:e.g. width of the Gaussian

Sensible ‚upgrade‘:p(x,xi) -> p(x,xi) / σito incorporate uncertainties

Adjust to give a reasonable ‚version‘of χ2/#points

χ2/#points

# en

trie

s

19

First results

First results look promising Scale linear within ±1% Small deviations from ‘perfection‘: ~1% rise at low energies Nevertheless: seems to be the way to go!

19

20

Summary and Conclusions

New global cell energy density weights for MC09

– Calculated using AntiKt 4 Topo Jets Numerical Inversion factors will be use to recover JES

– Calculated for all Jet Collections with fitting fuctions Still TO DO:

– Check energy density weighs and JES factors with other samples

– Store them into database to be use in standard production

– Document how the weights and factors have been calculated

– Understand origin of fluctuations

– Examine different techniques / parameters for smoothing• Find technical way to implement into JetCalibTools• TH2‘s from database already used in local hadron calibration• Could encode TGraphs in TH2‘s to simplify Athena modifications

21

BACK-UP

22

Reconstruction Details: JetSampling calculation

Package Tag

Athena Release 15.3.0

Reconstruction/Jet/JetEvent JetEvent-01-01-27 (from release)

Reconstruction/Jet/JetEventAthenaPool JetEventAthenaPool-00-00-14 (from release)

Reconstruction/Jet/JetEventTPCnv JetEventTPCnv-00-00-50 (from release)

Reconstruction/Jet/JetUtils Update to JetUtils-01-01-11

Reconstruction/Jet/JetRecTools JetRecTools-00-01-23 (from release)

Database/AtlasSTLAddReflex AtlasSTLAddReflex-00-00-18 (from release)

Reconstruction/Jet/JetCalib Update to JetCalib-00-04-20

23

OLD weights: Cone7Tower

24

Linearity and resolution without Num. Inversion

Cone7Tower

NO num. inversion