170 likes | 261 Views
Object Collection Cross Cleaning using PAT. SudhirMalik Fermilab/University of Nebraska-Lincoln. PAT Tutorial – CERN – 10 March 2010. Object Reconstruction with CMS. Physics Objects and POGs. Physics Objects are reconstructed independently by different POGs Muon Jet Electron Photon
E N D
Object Collection Cross Cleaning using PAT SudhirMalik Fermilab/University of Nebraska-Lincoln PAT Tutorial – CERN – 10 March 2010
Physics Objects and POGs • Physics Objects are reconstructed independently by different POGs • Muon • Jet • Electron • Photon • Tau • This can/will lead to ambiguities across different collections • Muon • Jet • Electron • Photon • Taus ? • Jet • Electron, Photon, Muon • Photon • Electron • Jets ? ? ? ?
Typical Ambiguities • Example 1: isolated Electron versus Jet: • Electron: cluster in the ECAL + pointing track in tracker • Calorimeter Jet: clustered energies in the ECAL + HCAL • The ECAL cluster will also show up reconstructed as a calorimeter jet • Example 2: Electron versus Photon: • Electron: cluster in the ECAL + pointing track in tracker • Photon: cluster in the ECAL + w/o track in tracker (BUT there can be conversions) • On the pure calorimeter level electrons show up as photons
Typical Concerns • Example 1: isolated Electron versus Jet: • Not an issue for a inclusive QCD jets analysis (only considers single object type) • More an issue for a ttbar + jets analysis in the semi-leptonic channel with electrons • Double counting of objects • Double counting of energy in the detector Cross Cleaning (object disambiguation) matters, if you want to interpret parts or the whole event
A Word of Caution Cross Cleaning and its exact configuration is a very analysis dependent procedure • Different aims for analyses and attitudes of PAGs : • QCD Jets • Inclusive V-Boson production • V+Jets • Top • Top+Jets • SUSY+Exotica
Concepts for Disambiguation • We have a reference collection and a set of test collections • E.g. test=Jets – reference=electrons • We have a matching criterion • E.g. geometrical matching (by DeltaR) • We might want to apply a pre-selections to the reference collection • E.g. electrons should be isolated
Cleaning in PAT SWGuidePATWorkflow • End of the Food Chain • After full PAT-ification • python/cleaningLayer1 8
Cross Cleaning in the PAT Workflow • Cross Cleaning is the last step in the patTuple creation chain: • It's the most high-level analysis operation • Input are the 'highest analysis objects' up to then • If I don't want to apply cross cleaning I just stop before • Or use the removeCleaning(process) tool • Even when cross cleaning is applied NO information and NO objects lost!
Algorithmic Procedure (PAT Default) Muon • Keep as it is • Clean if overlap (within ∆R<0.3) withMuon • Clean if same SuperClusterSeed as Electrons • Clean if overlap (within ∆R<0.3) with Muon • Clean if overlap (within ∆R<0.3) with Electron • Clean if overlap (within ∆R<0.5) with Muon, Photon, Electron • Clean if overlap (within ∆R<0.3) with track-isolated Electron • Clean if overlap (within ∆R<0.5) with Tau Electron Photon Tau Jet Here the order Matters: As some cleanPatCandidateshave other cleaned collections as input!
Configuration of Electrons edm::InputTagsrccommon input source std::stringpreselectionpotential pre-selection edm::PSetcheckOverlapsconfig. of overlap checking std::stringfinalCutpotential final selection src input source for test collection algorithm 'bySupperClusterSeed' algorithm preselectionpreselection for ref collection checkReco check for common reco components paircut apply cut on pairing (4-vector kin.)1) requireNo...drop overlaps from this collection 1) for potential cuts in pairCut have a look at PatUtils/interface/PatDiObjectProxy.h 11
Checking & Handling of Overlaps /CMSSW/DataFormats/PatCandidates/interface/PATObject.h The return values are of type reco::Candidate, but as you know of what type the test collections were you can also access specific information via dynamic_cast 14
Summary • Cross Cleaning is an issue for analyses that plan to interpret whole events • It is very analysis dependent and people are very cautious about it • PAT provides a very flexible frame for a standardized way of Cross Cleaning • The way how to apply cross cleaning • Which objects to consider for cross cleaning • In what order to apply cross cleaning • There is NO LOSS of information with it (unless stated explicitly) • In contrary we have a gain of information (selectedPatCandidate+more) 16
Exercises Exercise 1: • cvs up r patTutorial_mar10_module3a PhysicsTools/PatExamples • Take the cleanPatJetcollection as it comes with the PAT default and analyze it with the PatBasicAnalyer. Check how many overlaps the jets have with electrons in general and with track-isolated electrons. • Exercise 2: • Create a patTuple that contains both selected &cleanPatCandidates.Modify the default behavior for jets such, that it discards jets that overlap with isolated electrons. • Make use of PhysicsTools/PatExamples/test/analyzePatBasics_modified_cfg.py to compare the two Exercise 3: • Configure the module for jet cleaning such that it discards all jets that overlap with anyelectron or photon with a pt>20GeV. Compare the two cleaned jet collections using the PatBasicAnalyzer. 17