Bremsstrahlung effect in H 4 l analysis 13

Bremsstrahlung effect in H 4 l analysis (13. 0. 30) Fany Dudziak, LAL Lydia Fayard, RD Schaffer, Ilija Vukotic Higgs WG meeting, ATLAS CERN May 14 2008 1

Outlook • First goal: identify inefficiency sources in electron reconstruction, especially at low Pt => look at brem. • Rates of brems, their energy and where they happen. • Brem effect in H 4 l =>event losses in: – the matching efficiency between generated and reconstructed electrons. – Is. EM trackmatch φ cut inefficiencies – Some variables sensitive to brem which could be usefull for the analysis. 2

Where the brems happen, their energy and their rates (1) • Electrons can emit a lot of brems before the cryostat which can be very energetic : Energy of the brems Number of brems per electron 25% have E>5 Ge. V Eγ • The number of brems emitted is correlated with the quantity of material before the calorimeter 3 η η

Where the brems happen, their energy and their rates (2) • To characterize the importance of the brem we have to know how much energy is radiated from the electron Definition : At the generation level : loop over the children (1 e & 1γ) of the electron coming from the Z and make the sum of the γ’s energy fraction of radiated energy ΣEγi/Ee(gen) In the center of the barrel few brems emitted and little energy is carried away For |η|>0. 6 the majority of energy is radiated. η 4

ΣEγi/Ee(gen) Where the brems happen, their energy and their rates (3) For BARREL (|η|<1. 5) We see that the brems happen in the regions with dense materials in the inner detectors The effect on the reconstruction depend on the place it happens and the quantity of energy taken out the electron. ΣEγi/Ee(gen) R(last Brem) For End Cap (|η|>1. 5) R(last 5 Brem)

Point : • In case of ΣEγi/Ee(gen) 1 , two possibilities: - most of energy is radiated early, in the trackers: we reconstruct poorly the track of electron (few hits) (worse case : missing track, electron reconstructed as photon) - most of energy is radiated near cryostat : we have a track but cluster’s shape is enlarged in φ. γ cluster Bad track 6

Inefficiency of the matching between generated and reconstructed electrons and brem (1). • Efficiencies in 13. 0. 30: • Matching in a cone ΔR<0. 2 • Inefficiency about 14% Those events have emitted most of their energy as photons : « big » brems are in end cap We can look at the photon container … ΣEγi/Ee(gen) – Efficiency in barrel region : >90% – Efficiency in End Cap region : falling to less than 80% 7 η

Inefficiency of the matching between generated and reconstructed electrons and brem (2). • When there is no matching between a true and a reconstructed electron (with ΔR<0. 2), I have searched for a photon in the photon container : Eγ(reco)/Ee(gen) Impact on the efficiency : - efficiency of e matching in 0. 2 : 85. 8% - efficiency adding photons matched in a 0. 2 cone : 89. 9% - efficiency adding photons matched in a 0. 3 cone : 89. 9% 4% 8

Some variables sensitive to brem (1) • At the reconstruction level we define 3 variables sensitive to brem 1) Width in φ: RMS of the cell’s φ in the 2 nd sampling i cell’s index in the middle RMS(φ) for the barrel RMS and its tails increase with the amount of brem RMS(φ) 9

Some variables sensitive to brem (2) Δ (2) the difference between the cluster position and the track position: φ( rk l-t ca ) Δφ(cal-trk) = φ(cluster)-φ(track extrapolated to the middle) The sign of the particle is taken into account Δφ(cal-trk) - Tails are important for high brem - Mean is slightly offset. - Both effects larger for low Pt. 10

Some variables sensitive to brem (3) Δφ(cal-trk) (3) the fraction of energy deposited in the presampler increases as the shower begins earlier. E(PS)/Ecluster 11

Observed inefficiency of Is. EM trackmatch φ cut (1) Tight Is. EM cut is cutting the tails of Δφ(cal-trk) Track. Matchφ is cuting the Δφ(cal-trk) distribution at -0. 02 value for every Pt and η (bit : 0 x 3 F 7 FFF) Efficiencies after Ox. F 7 FFF + … 1% 7% η η efficiency loss mainly for lower Pt and uniform in η Pt(Me. V) 12

Observed inefficiency of Is. EM trackmatch φ cut (2) • Δφ depends on η and Pt. Δφ(cal-trk) • We plot Δφ for all Pt-η bins used in Is. EM cuts: Points are the mean of the Δφ(caltrk) distribution and barres are the rms – For low Pt the distribution is spread and the means are higher. – In order to keep electrons with brem, one has to optimize this cut in eta-pt bins. What would be the effect on the energy reconstruction What would be the effect on the background rejection 13 Pt

Observed inefficiency of Is. EM trackmatch φ cut (3) Energy reconstruction: First look: the energy reconstruction does not depend on Δφ(cal-trk). It is just a bit better for higher values Zbb Background: Zbb H 4 l track. Matchφ track. Matchη 5. 05%e cut 5. 25% e cut track. Match. E/P track. Matchη 5. 50% e cut 4. 47% e cut If we widen the Δφ(cal-trk) acceptance we will gain at most 5. 05% Zbb background. signal Zbb background 14 Δφ(cal-trk)

Conclusion Brem’s effect in 13. 0. 30: • 4% of electrons found as photons – Probably because of strong brem – Could we combine the electron and photon containers? – Wait for a treatement of brem in track reconstruction. • Tight track. Match φ cut rejects 5% of electron, mainly in region of low Pt – Need to optimize the corresponding cut in Pt and η. – A first look indicates : • The per electron rate increase for the Zbb background is about 5% • Expect the ZZ background increase less due to higher Pt electrons. • We can use other variables to characterize the brem. We have to explore how best to use them to improve the reconstruction. • All this has to be refined with generic samples (Z->ee …) and discussed within the egamma group. 15

Now • In the case of no matching with electron nor photon look in the LAr. Cluster. Container (ESD) if we can find a cluster where the e/γ was expected. – First look indicates that we can find something – Back navigation allows to look 1 AOD with 1 ESD 1 sample of 1000 H 4 l ev used. 1948 electron generated (|η|<2. 47) 1703 (e) + 78 (γ) reco found. ~10% lost (167 e) Very first look : we find a cluster in some cases (but with the 4 th vector of the gen elec, has to be extrapolated to the calo) 16

Now Eclus/Egen Lot of things to be understood ……. 17