180 likes | 341 Vues
This study investigates the search for RPV SUSY through LLE coupling, presenting initial results from data analysis. Utilizing advanced neural networks, we optimize tau identification in hadronic final states under various decay modes. With a dataset of 246.7 pb⁻¹, we assess backgrounds, employing neural networks designed for tau candidates and implementing specific cuts to mitigate electron and muon contamination. Our methodologies offer a preliminary estimate of NN efficiencies and the underlying physics of the decay path, while also addressing discrepancies between data and simulation in terms of efficiencies and backgrounds.
E N D
Search for RPV SUSY through the LLE coupling Anne-Catherine Le Bihan / François Charles RPV final sate Tau-Id with neural networks NN efficiencies on data Data selection & efficiencies First result on 246,7 pb-1
RPV final state : decay via coupling Final state : 2 LSP decays with coupling : • 2 taus + 2 electrons • 3 taus + 1 electron • 4 taus => look for 2 isolated electrons and at least one hadronic tau AC Le Bihan
Reminder : tau identification in hadronic final states • 3 main backgrounds: • QCD background : make use of 3 neural networks • 2 NN designed for taus with one associated track (w/ or wo/ EM3 subcluster) • 1 NN designed for taus with at least 2 associated tracks • (last version - Arnaud Gay http://ireswww.in2p3.fr/ires/recherche/dzero/tauid.html) • electrons : veto on candidates with an electron matched in an 0.3 cone • + use specific NN trained on Z->ee • muons : veto on candidates with a medium muon matched in an 0.3 cone • + cut on E(tau)-ECH_5x5(trk)/pT(trk) > 0.7 ( p14 certification - D0 note 4453 ) AC Le Bihan
Z-> -> enriched sample Enriched sample : muloose CSG skim processed with muon isolation filter (>=p14.05) : select events with one isolated muon and 1 tau candidate of opposite sign, (mu,tau)>2.7 + kinematical cuts ET(tower1)+ET(tower2)/ET Background sample : select events with a muon in a jet and a tau candidate of same sign (mu,tau)>0.7 Try to estimate the Z-> content by fitting the profile distribution (for all hadronic modes). AC Le Bihan
tracks not attached to a tau in an 0.5 cone / all tracks Z-> -> enriched sample Check fit on other NN input variables : Cluster isolation : pT(R=05)-pT(R=0.3)/pT(R=0.3) AC Le Bihan
Electron and muon contamination for and decay modes Before cutting on : E(tau)-ECH_5x5(trk)/pT(trk) > 0.7 Before cutting on NN(Z->ee) : electrons faking hadronic taus AC Le Bihan
tau estimate (decay mode ) : 405 tau estimate (decay modes and ) : 848 tau content and AC Le Bihan
First estimate of NN efficienciesand profile cluster isolation NN( ) > 0.8 or NN( ) > 0.8 : MC : 541 events data : 410 events (data) : 48 % (MC) : 64 % (data)/ (MC) : 76 % (QCD bkg) : 2.2 % AC Le Bihan
Remaining electron contamination Use e+tau invariant mass to check the remaining electron contamination : e loose : emfrac >0.9 iso<0.15 e tight : emfrac >0.9 iso<0.15 trackmatchchi2prob#-1 HMX8<50 likelihood >0.3 • How many loose electrons give a tau ? => match loose electron to tau e(tight) + e(loose) invariant mass e(tight)+tau invariant mass => 7% of loose electrons give a tau AC Le Bihan
Data sample - cut flow Preselection : 2EM CSG skim, fixed with d0correct v6 - v6-a Trigger : single EM and Di-EM Bad calo runs, bad met and ring of fire lumi-blocks removed Dupplicated events and bad sam files removed Luminosity : 246, 7 pb-1 • Cut flow : 1) 2 electrons (likelihood >0.3, HMX8<50), M(ee) > 18 Gev/c2 1) at least 1 hadronic tau (decay mode or ) outside the ICD region 3) M(ee) (81,101) 4) met/sqrt(set) > 1.5 AC Le Bihan
RPV signal Points generated with susygen v03-00-43 , 133 = 0.003: • can be lighter than : • -> : 100 % • -> : 100 % • => higher tau final state multiplicity AC Le Bihan
trigger efficiency folded into MC Use allmost all EM and Di-EM triggers , v8-v12 Use Ulla’s method to determine the turn-on function : -> compare MC/data for events with 2 electrons, 15<M(ee)<60, met<10 Freq((pT-12.2)/sqrt(pT*0.66)) pT of leading electron (Gev/c) AC Le Bihan
electron data/MC efficiencies EC-EC CC use tag and probe method on tight electron + track invariant mass: _CC(data) : 81 % _CC(MC): 86 % _EC(data) : 67 % _EC(MC): 93% Like lihood efficiency : use tag and probe method on tight electron + loose electron invariant mass: _CC(data) :89% _CC(MC):97% _EC(data) :68% _EC(MC):65% AC Le Bihan
Cut 1 - 2 electrons : M(ee) Use Anne-Marie’s smearing : (E) =0.047*E Multi-jet QCD background evaluated by inverting the cut on tau-electron NN : select events with 2 electrons with NN> 0.05 (more tau ~ jet like) AC Le Bihan
Cut 1 - 2 electrons - met/sqrt(set) After correction of the electron smearing : discrepancy needs to be understood … AC Le Bihan
Data / backgrounds - cuts 1,2,3,4 AC Le Bihan
RPV signal - cut 3 M(ee) met/sqrt(set) AC Le Bihan
Conclusion and plans • first estimate of NN efficiencies for hadronic taus on p14 data • added cut to remove electron and muon contamination • Preselected data agree reasonably well with the expected background • check signal cuts on points above the LEP limit : m( )>103 GeV/c2 • understand better MC/data differences for NN efficiencies • understand the met difference MC/data • use matrix method to cross-check the QCD background estimate AC Le Bihan