1. MissingET and Z tt with first data III Physics Workshop
di Atlas Italia
S. Resconi

2. Attivita’ italiana su EtMiss e Tau:
Milano (D. Cavalli, S.R.) sviluppa e mantiene il MissingET package in ATLAS:
Migliorie degli algoritmi per garantire:
Flessibilita’ nella ricostruzione e calibrazione
robustezza con i primi dati
Soddisfacenti performance su diversi canali di fisica
Soppressione delle sorgenti di Fake EtMiss
Risultati documentati nella EtMiss CSC Note (editor D. Cavalli)
S.R co-convenor della “Jet/ETMiss Data Preparation task force”
Frascati (M. Testa) studi di EtMiss in eventi Z/W
Tau:
Intensa attivita’ del gruppo di Milano su ricostruzione e identificazione dei Tau in passato, ora concentrata sullo studio di canali con Tau nei primi dati:
Ztt (D.Cavalli, C. Pizio, W. Davey)
W tn (A. Andreazza, L. Dell’Asta) attivita’ che sta iniziando ora 2

3. EtMiss: importance for physics
A very good EtMiss measurement is a crucial requirement for the study of many physics channels in ATLAS.
EtMiss is the signature for many physics channels:
Wln, Ztt, top decays...
SM Higgs (VBF H tt, ttH  tt )
MSSM Higgs (A/H  tt, H±  tn)
reconstruct the invariant tt mass from the two EtMiss components
SUSY
Large EtMiss signature (LSP)

4. Sum of energy of all particles seen in the detector
EtMiss measurement p pi
Sp=0  Spi=0
Transverse Missing Energy:
EtMiss= ExMiss2+EyMiss2
ExMiss =-SEx
EyMiss =-SEy
SumET = SET
Sum of energy of all particles seen in the detector
SEx e SEy
EtMiss is event quantity
all detectors contribute
It is calculated from all significant signals
All calorimeter signals
Calorimeter signals in physics objects
Calorimeter signals not used in physics objects
Muons
Tracks in regions where calorimeter/muon spectrometer are inefficient
Correction for energy lost in dead materials

5. Fake EtMiss in early data: event cleaning
EtMiss is due to non interacting particles in detector
(neutrinos, LSP)  True EtMiss
BUT also it is due to:
Problems in detector (dead, hot, noisy channels, problems in HV sectors…)
Other particles arriving at the detector (cosmics, beam-gas, beam-halo...)
Electronic noise, pileup noise
Energy lost in dead materials (cracks, cryostats…)
Particles outside the detector acceptance (calorimeter coverage: |h|5, muon spectrometer coverage: |h|2.7)
Many other effects
 Fake EtMiss
First require detailed understanding of instrumental Etmiss sources  event cleaning
 See effect of Etmiss cleaning in D0 and CDF
Then understand other source of Fake
EtMiss (missing/fake muons, jets in cracks…) D0
CDF

6. EtMiss reconstruction in ATLAS
Event Data Model and AlgTools structure provide high level of modularity:
Event Data Model (MissingETEvent package):
 EDM is made of simple and general classes that use enumerators to specialize data:
 is suitable for MissingET data:
basic unit is made of the 3-vector:
(m_exMiss, m_eyMiss, m_etSum)
but can have many different sources :
calorimeters, muons, truth, DM regions…
 permits to minimize the number of data classes
 gives flexibility to the architecture, easy to be extended:
to add a MET new object just add a new source, no EDM changes !
AlgTools structure and flow (MissingET package):
 implemented two main Algorithms: METAlg and METRefAlg
to handle two main approaches for EtMiss calculation and calibration:
 Global Calibration (MET_Final) and Refined Calibration (MET_RefFinal)
 AlgTools configuration in /python/METGetter.py and METRefGetters.py
 each AlgTool is responsible to create/fill and record in StoreGate one MET object
 this allows to implement a step by step procedure of increasing complexity
for the commissioning of the EtMiss
 different levels of “descoping” for EtMiss reconstruction already tested in the
reprocessings of the cosmics and single beam data during 08/09

7. Flow of MET Tools with first data
With the higher level of descoping
very basic cell-based EtMiss
Calculation applying two different
noise suppressions:
(1) from all Cells with |E|>2 noise
(2) from all Cells inside TopoClusters
 Provided also the possibility to
mask bad channels
 No calibration
(1)
CaloCellContainer
(2)
CaloTopoClusters
Cell Loop
METCaloCellTool
METtopoObjTool
Loop on TC and
backnavigate to
TopoCells
2 noise cut
METSelectorTool
CaloCell
TopoCell
METCaloTool
MET_Topo
MET_Base
METCellMaskTool
NEW
NEW
MET_Base_Mask
MET_Topo_Mask
METCaloTool
NEW
• LArBadChanTool/Masker, TileBadChanTool combined in CaloBadChanTool
 access channel quality and mask channels
• METCellMaskTool provides an interfaces to these tools to mask cells in MET reco

8. Masking of bad channels in MissingET
Some results on single beam data: run 87764/physics_L1Calo/ with rel
PEMB
EMB
PEMEC
EMEC
TILE
HEC
FCAL
400 20
(MET_BASE – MET_BASE_MASK) vs MET_BASE_MASK
MET_BASE – MET_BASE_MASK
MET_BASE
MET_TOPO
After masking all bad channels:
 big effect on small MET and SUMET values
 few events particularly affected for MET > 20 GeV
Tile Calo gives the main contribution (probably needed a more refined classification
of “affected” channels in Tile)
 Masking of bad channels should be done at Calorimeter Cell level to provide coherent reconstruction
of the entire event, but if needed possibility to mask only in MET reconstruction
Control plots in different subdetector regions

9. MissingET with random trigger
• Significant improvements in events with random triggers in cosmics in last year:
improved bad-channel data base and masking of bad channels at calorimeter level
improved energy reconstruction
improved noise description
SumET= scalar sum of Cell ET with |E|> 2 noise
 should be gaussian centered to zero in random trigger evts ( expected)
 improvements from 1° reprocessing
to last reprocessing
two algorithms:
METbase: from all Cells with |E|> 2 noise
METtopo: from all Cells in TopoClusters  provides better perfomance !

10. Calorimeter based MET reconstruction and global calibration
(1)
CaloCellContainer
(2)
CaloTopoClusters
(3)
CaloCalTopoClusters
Cell Loop
METCaloCellTool
METtopoObjTool
MET_TopoObj
Loop on TC and
backnavigate to
TopoCells
MET_LocHadTopoObj
2 noise cut
METSelectorTool
CaloCell
TopoCell
METCaloTool
MET_Topo
NO calib
MET_Base
NO calib
H1 calib
MET_Calib
METCalibTool
MET_CorrTopo
H1 calib
 Cell weights are applied for EtMiss global calibration:
H1-style calibration:
cell weights, depending on cell energy density E/V, on |h| and on calorimeter region, determined optimising the jet resolution
LocalHadron calibration:
cell weights obtained using MC calibration hits info: energy deposition (active and inactive parts) in detector and dead material.
can calculate EtMiss from calibrated cells and from calibrated Topoclusters: similar results
MET_LocHadTopo
LocHad calib

11. Final MET reconstruction
METFinalTool
MET_Final
MET_CorrTopo
MET_MuonBoy
MET_Cryo
To calculate the Final MET should be added also:
MET_Cryo = Correction for energy lost in cryostat between EM and Had calorimeters,
Sum Ex and Ey cryostat correction from each jet:
 MET_MuonBoy = Muon term, calculated adding contribution from:
isolated muons: use combined tracks plus CaloMuons MuTag tracks to take into account also muons in crack regions respectively: |h| < 0.1, 1< |h| < 1.3
To avoid energy double counting the muon energy deposited in calorimeter is not added in the Final MET calculation
non-isolated muons: using combined reconstructed muons but taking the momentum from the external muon spectrometer only
For first data a more robust approach is forseen: no separation between isolated and non-isol, and use for ALL muons the combined reconstructed muons taking the momentum from the external muon spectrometer only (no problem of double counting in calos)

12. < 1 pb-1 data: in-situ EtMiss validation with minbias evts
control sample to test EtMiss resolution up to SumET~200 GeV
no MC Truth needed to build the resolution curve  No MET expected in minbias events
 (minbias) = 80 mb
 Resolution curve done with 100k evts
Final Ex(y)miss Resol = p0 *  SumET
METbase: from all Cells with |E|> 2 noise
METtopo: from all Cells inside TopoClusters provides better perfomance !

13. < 1 pb-1 data: in-situ EtMiss validation with QCD di-jet evts
 control sample to test EtMiss resolution up to SumET >200 GeV
 J4 pt(jet)=140 – 280 GeV,  (j4) = 150 nb
 Resolution curve done with 250k evts
Final Ex(y)miss Resol = p0 *  SumET
Reconstruction and calibration steps  resolution improves from MET_Topo to MET_Final:
MET_Topo  p0=0.56 from all cells inside TopoClusters (no calib)
MET_TopoCorr  p0=0.54 apply H1 calibration to all cells inside TopoClusters
MET_Final  p0= MET_TopoCorr + MET_MuonBoy + MET_Cryo
MET_LocHadTopo  p0=0.55 apply LocHad calib to all cells inside TopoClusters

14. Refined MET Calibration
implementation in Athena is based on an association map
between reconstructed objects and their constituent CaloCells and/or TopoClusters.
 the association map allows to remove overlaps at cell level
The set of METRefTools which access/fill the association map for each reconstructed object:
METRefEleTool
METRefTauTool
METRefJetTool
METRefMuoTool
METRefCellOutTool
Each RefTool :
loop on identified Reco Objects collections
for each Reco Obj backnavigate to its CaloCells
fill the association map with:
*CaloCell, *RecoObj, CaloCell weight
calculate METRef_obj
METRefCellOutTool :
loop on all CaloCells in TopoClusters not associated to reco object
fill the association map with:
*CaloCell, *TopoCl, CaloCell weight
calculate METRef_CellOut
METRef_Ele
METRef_
METRef_Tau
METRef_BJet
METRef_Jet
METRef_Muo
METRef_CellOut + + + + + +
+ MET_MuonBoy + MET_Cryo = METRef_Final
cell weights depend on the type of the reconstructed object (e/, , b-jet, jet, m …)
particle identification driven by MissingET package jobOptions.
 each contribution is individually available in ESD/AOD, degrees of freedom in physics analysis

15. 10 pb-1 data: in-situ EtMiss validation with Zll evts
Possibility to validate EtMiss performance without MC Truth info  no MET expected
Zee:  =1.4 nb, Analysis eff = 25%, in 10 pb-1  N = 3500 evts expected
Diagnostic plot of MET vs dilepton pt projected along perpendicular axis:
 checks calibration and scale of MET and should be flat at EtMiss=0
 Best calibration is the Refined Calibration and Local Hadron Calibration
 A negative offset up to ~ 3 GeV suggests some problems that is under investigation…
The perpendicular axis is defined by the vectorial sum of the 2 leptons momenta.
The parallel axis is placed at p/2 to the perpendicular axis.
The perpendicular axis is more sensitive to the balance between the electons and the hadronic recoil

16. 10 pb-1 data: in-situ EtMiss validation with Zll evts
…in particular splitting the diagnostic plot by jet multiplicity suggests miscalibration of
low energy deposits:
Worse negative offset for Zee events with njet=0, this means that in the Refined Calibration the term calculated from all cells outside reconstructed objects but inside TopoCluster (MET_CellOut) need an improved calibration
 still under investigation….

17. Determinazione “in Situ” della scala e della risoluzione
in ETMiss (lavoro di Marianna Testa, LNF)
Motivazioni:
Correzioni del MonteCarlo
Sistematiche sull' efficienza dei tagli e sull'accettanza in diversi canali (W® ln,...)
Sistematica dovuta alle “shape” del MonteCarlo per il conteggio dei segnali
Strategia:
Uso di Z® mm (ETmiss ~ 0) per valutare la scala e la risoluzione in ETmiss.
Decomposizione di ETmiss lungo le direzioni perpendicolare e parallela allo Z e al W
Fit gaussiani delle proiezioni di ETMiss -ETMissTruth (ETMiss) per il W (Z) in bin
di Sum ET per tenere conto del diverso rinculo adronico di W e Z

18. per la componente parallela. Buon accordo tra W e Z.
(lavoro di Marianna Testa, LNF)
Bias di ~3 GeV per la componente perpendicolare di Etmiss, non presente
per la componente parallela. Buon accordo tra W e Z.
Componente perpendicolare
Componente parallela
Componente parallela
Buon accordo tra la risoluzioni
in ETMiss per eventi
W ® mn e Z® mm, in entrambe le
proiezioni.
Per Wmn usata direttamente l’informazione
della direzione Truth del W

19. 100 pb-1 data: in-situ EtMiss validation with Ztt  lep-had evts
Z    lepton-hadron analysis:
developed by Milano group: D. Cavalli, C. Pizio in close collaboration with Freiburg group.
 First data in this channel can be used to:
determine the Tau scale from the reconstruction of the visible mass (lept, t-jet),
determine the EtMiss scale from the reconstruction of the invariant tt mass
determine the Tau-jet Identification Efficiency
measure the cross-section
 Analysis at 10 TeV and L= 100pb-1:
Select Z  lepton-hadron strict cuts applied to have low level of backgrounds
Main backgrounds: QCD,Wmn,Wen,Wtn, (tt,Zee,Zmm)
Estimate background in-situ using same sign events (SS events)
 signal events have opposite sign lepton and t-jet (OS events)

20. Analysis Method
Two separated analysis  use the invariant mass to tune the EtMiss scale and the visible mass to tune the tau scale.
 This analysis is now implemented in the Z/W Benchmark package
Select max pT lepton in the event
pT>10 GeV (15)
Use single ele/m trigger
Basic cut flow:
- ETMiss>20 GeV
-mandatory for MET scale determination
-helpful against QCD and Zll
- Lept – ETMiss Transverse Mass (mT) < 30 (50) Gev
- SET < 400GeV
5. Separate OS evts from SS events
Signal: only OS evts,
Backgds: OS and SS with similar probabilitybackground contribution
can be estimated in-situ using SS events
6. Subtract SS from OS evts
A correction is needed for W background where OS/SS≈1.5
 correction factor determined in situ (cfr. ATL-PHYS-INT )
3. Invariant/Visible mass reconstruction:
use e/m candidate and t-jet candidates
7.  ETMiss scale determination
from reconstruted invariant mass
 t scale determination
from reconstructed visible mass
4. Second cut flow:
- 1. < Dj (Lept – t-jet) < 3.1 (2.8)
- Invariant Mass tt >0

21. Subtraction of backgrounds in-situ using SS events
Main backgrounds are QCD, W+jets:
QCD: same probability for OS and SS
W+jets: OS/SS=1.5
in pp two production channels (qq’->SS/OS, qg->OS)
Procedure to evaluate number of W OS evts:
Do not apply the mTlep-METcut
From mTlep-MET distribution at the end of cuts :
- Evaluate from data the number of W SS (NSSCONT) and OS events in the W control region (RSOS = OS/SS):
 for 50<mTlep-MET (GeV)<100 only W events are collected (other evts/W evts= 2.5%)
Get from MC the fraction of W evts with mTlep-MET < 30 GeV in signal region respect to control (RCONT= from MC )
compare different MC…
3) Evaluate the number of W events OS expected in Signal region: NOSSignal region=RCONT*RSOS* NSSCONT
W+jets
Signal
region
W control
region

22. Z tt: tune Missing ET Scale from invariant m
OS all
OS Signal
SS all 1 2
OS-SS
tt invariant mass distributions:
1) OS (signal + background), SS (signal + background), NO QCD
2) OS – SS
In 100pb-1  200 Signal evts S/B=22
S/B ~ 5 taking into account QCD background (still preliminar…) QCD can be suppressed optimizing the lepton isolation cut

23. Z tt: tune Missing ET Scale from invariant m
Determination of the EtMiss scale with invariant m :
in 100 pb-1 invariant m mass reconstructed with an error of less then 1 GeV (0.8 GeV)
taking into account only the statistical error the EtMiss scale could be determined with a precision of 3 %
taking into account also systematic effects, due to subtraction of the SS events and the stability of the fit, EtMiss scale could be determined with a precision of 8 %
preliminar
+3
+1
-1
-3
Inv mass vs ETMiss Scale
OS Signal evts
OS-SS Signal+Backgd evts
(the one that we will have in real data !)
NO QCD included

24. Z tt: tune Tau scale from visible mass
Determination of the tau scale with visible mass analysis:
in 100 pb-1  400 signal events S/B=5
visible mass can be reconstructed with an error of less then 1 GeV
taking into account only the statistical error the Tau-energy scale could be determined with a precision < 3 %  also systematic error due to subtraction of the SS events and the stability of the fit has to be taken into account
Visible mass vs Etau Scale OS Signal evts
+3
+1
-1
-3

25. BACK-UP SLIDES

26. MissingET reconstruction on AOD
Request from physics: possibility to rerun quickly MET reconstruction changing calibration and/or particle identification.
New code is ready in rel :
Provides the possibility to rerun refined calibration on AOD
New design by P. Loch permits to use same METRefTool for ESD and AOD reco configuring themselves with processors according to cell or cluster use
Which constituents are used at AOD level ?
constituent TopoClusters of taus and jets (no more cell connection on AOD)
constituent CaloCells of electons/photons/muons (available on AOD)
How to apply overlap removal at AOD level ?
unfortunately cannot use only TopoClusters because electons/photons/muons are not built on TopoClusters
new method to associate CaloCells to TopoClusters
What is missing at AOD level ?
To have all possible calibrations available also at AOD level:
needed to use all 3 signal states (H1, LocHad, EmScale) signal states for TopoClusters, H1 signal states is a recent development and still under test.

27. validation of MissingET reconstruction on AOD
MET reco on ESD (nightlies for rel )
MET reco on AOD (nightlies for rel )
mc PythiaZtautau
MET_RefEle
MET_RefGamma
MET_RefJet
MET_CellOut
Some differences expected in AOD reco :
• Different calibration for taus/jets/cellOut:
 on ESD H1 applyed at cell level, on AOD locHad at TopoCluster level
• MET_RefJet (include also taus)
MET_RefFinal

28. Primary Performance DPD
For EtMiss performance studies with first data all kinds of performace DPDs with all CaloCells can be used
For Ztautau analysis:
Single Egamma/Muon performance DPDs with very first data
Checked PERFORMANCE DPD in Ztautau analysis (by C. Pizio):
all quantities needed for the analysis are there
all MissingET quantities are there
Need pT lepton threshold at 10 GeV (done! ) isolation cut (done!)

29. Tau identification with first data
Many discussions inside tau WG to define a set of “safe variables” on to be used for tau-ID cuts with first data instead of using the Likelihood (pT dependent)
Two approaces: calo-based and calo+track-based
Work to optimize the cuts for 3 selections: loose (eff=0.7), medium(eff=0.5), tight (eff=0.3) on signal sample (Ztt and Att) and di-jets background samples (J1-J5)

30. Analysis with safe tau variables
BACK-UP SLIDES

31. Topological clusters and noise suppression
the best noise suppression in calorimeters is achieved if EtMiss is calculated from cells in Topoclusters 4/2/0:
Topological clusters (Topoclusters) are built from all Calorimeters
• by grouping cells which are topological neighbours with 3 Signal/Noise
Thresholds:
|E/noise| > Tseed (default Tseed = 4); cells above this threshold initiate
or expand a cluster and are asked for their neighbours
|E/noise| > Tneighbour (default Tneighbour = 2); cells above this threshold
expand a cluster and are asked for their neighbours
|E/noise| > Tcell (default Tcell = 0); only cells above this threshold are used
• with noise being either electronic-noise or electronic-noise+ pile-up-noise from CaloNoiseTool

32. Local Hadon Calibration
The local hadronic calibration for the ATLAS calorimeter system
is a tool to get calibrated jets at particle level for any jet algorithm.
The procedure is based on detailed Geant4 simulations providing information on energy deposits in all parts of the ATLAS detector.
Calibration starts from topological clusters reconstructed and calibrated at the electromagnetic scale.
First classification step tags clusters as mainly electromagnetic, or hadronic depending on cluster shape variables:
Clusters which are identified as being electromagnetic are kept at the electromagnetic scale.
Hadronic clusters receive cell weights to correct for the invisible energy deposits of hadrons.
Next steps are out-of-cluster corrections for lost energy deposited in calorimeter cells outside of reconstructed clusters and dead material corrections for energy deposited outside of active calorimeter volumes.

33. Event Data Model (MissingETEvent) :
Enums for Source
of Signals:
Calo,
Calib,
Truth,
Muon,
Final,
Cryo,
Topo,
Ref,
ObjElectron
ObjMuon, ObjJet,
ObjIdTrk,
ObjMiniJet,
ObjRest,
ObjFinal
Enums for  Regions:
Central,
EndCap,
Forward
Enums for Truth Source:
Int = all int particles till abs(eta) <= 5
NonInt = non interacting particles
IntCentral = int particles in abs(eta) <= 3.2
IntFwd = int parts in Forward region
IntOutCover= int parts with abs(eta) > 5
Muons = truth muons ( all )
Enums for Calo Regions:
PEMB, EMB , PEMEC , EMEC ,
TILE, HEC, FCAL

35. tt Invariant Mass Reconstruction
m =  2(Elept+ E1 )(Et-jet+ E2)(1 - cos)
Elept,Et-jet =  decay products energies
j = decay products directions angle
E1, E2 = neutrinos systems energies
Assumptions:
m = 0
collinearity
Neutrinos systems energies are obtained solving the system:
Ex= (En1*u1)x + (En2*u2 )x
Ey= (En1*u1)y + (En2*u2)y
This system can not be always solved
Determinant has to not be zero (sinDf≠0)
 not back-to-bact lepton & t-jet
E1, E2 have to be > 0
Visible mass = m(lept, t-jet, ET)= (Plept +Pt +ET)2