50 %
50 %
Information about electronID

Published on January 20, 2008

Author: Pasquale

Source: authorstream.com

Electron Identification in ATLAS:  Electron Identification in ATLAS Introduction Calorimeter Reconstruction Inner Detector Reconstruction Combined Reconstruction Rejection/Efficiencies Performance in Rome What needs to be done to be more realistic Introduction:  Introduction E/ Identification important for lots of physics channels such as Higgs(SM, MSSM), new gauge bosons, extra dimensions, SUSY, W, top Need to cover the full range of pT from few GeV (B-physics) up to ~1TeV (new gauge bosons…) Examples: H  Most promising channel for Higgs if mH<130 GeV H  ZZ*  4e Most promising channel for Higgs if mH>130 GeV H  WW, ZZ  ejj, ee, eejj Interesting for heavy Higgs mH<1 TeV New extra gauge bosons: Z ’ ee, W ’  e tt production Rare decay modes of one top-quark Z  ee Important for e.m. calorimeter calibration + lots of other channels, e.g. Susy Calorimeter Energy measurement:  Calorimeter Energy measurement Design Goal Resolution ~10%/E up to 300 GeV Constant term < 1% Linearity better than 1% from few GeV up to ~TeV Layout 3 layers in depth main compartment 2nd layer ‘Small’ cells in eta in 1st layer Pre-sampler in <1.8 How do we measure the energy?:  How do we measure the energy? Cells calibrated at EM scale using optimal filtering Use longitudinal weights (eta dependent) Compromise between best resolution and best linearity Cluster reconstruction ‘default’: sliding window Topological clustering strips e 50GeV Middle Back Presampler LAr Calorimeter Upstream Material Upstream Losses Losses between PS and S1 Longitudinal Leakage Sliding window algorithm:  Sliding window algorithm ‘default’ clustering algo in athena for EM particle reconstruction Sum energy in depth in =0.0250.025 Cut cells artificially and share energy Look for ‘local’ maximum using sliding window Take candidate window Look at energy in window shifted by one cell to left/right, up/down, vertical and decide if you have local maximum Move window to next position EL E ER ED E E EU E>EL , E>ER E>EU E>ED Sliding window Clustering:  Sliding window Clustering Build up cluster in given window (37, 35, 55) Now use cell information, no longer towers Correct for S-shapes, eta/phi modulations, leakage outside the cluster… Corrections applied: Dependence SIZE (approx) S shape in eta, middle f(eta,Energy) Phi offset f(eta,Energy) S shape in eta, strips f(eta,Energy) E vs phi local modulation f(eta,Energy) 0.5% E vs eta local modulation f(Energy) 0.2% Gap correction f(eta) Longitudinal weights f(eta) 3-10% corrections derived from G4 single-e samples (Scott+Stathes) Longitudinal Weights: E(corr) = Scale(eta)*(Offset(eta)+W0(eta)*EPS+E1+E2+W3(eta)*E3) Latest iteration of longitudinal weights using 9.0.4 single electrons. Out-of-cone corrections aborbed in Scale(eta) Some corrections different for 5x5, 3x5 3x7 Topological clustering:  Topological clustering More ‘flexible’ than sliding window as it ‘adapts’ itself to the shower Cluster is built around a Seed Cell which has an energy above a certain threshold (the Seedthreshold) The neighbours of the Seed Cell are scanned for their energy and are added to the cluster if this energy is above the neighbour threshold. Then the neighbours of the neighbours are scanned and so on. The cuts, which are made for the seed and the neighbour, depend on the noise in each cell Similar cluster corrections needed as for sliding window S-shape in eta, Phi offset, E vs phi modulation corrections implemented but not applied yet eta Seed Cell Linearity Performance DC2 layout (10.0.1):  Linearity Performance DC2 layout (10.0.1) S. Snyder +1% -1% 1TeV eta eta eta eta Erec/Etrue Energy Resolution Comparison with TDR:  Energy Resolution Comparison with TDR EM Resolution is up to 25% worse than in the TDR G.Unal 10.0.1 TopoCluster(630) vs 3x7 :  10.0.1 TopoCluster(630) vs 3x7 100GeV topo 3x7 20GeV topo 3x7 Flores,Mellado,Quayle,Sau Lan Wu (topo and 3x7 recalculation of Long. Weights) topo Resolution s: similar to 3x7 Linearity: systematic shift: ~0.4%, maybe more over larger energy range. Needs to be understood. RMS: improved with TopoCluster. Expected since outliers more likely to be caught by TopoCluster. Needs to be evaluated in realistic environment. EM shower shape analysis:  EM shower shape analysis shower shape variables useful for e/jet (/jet) separation leakage into 1st sampling of had calo Transverse shower shapes in 2nd sampling Transverse shower shapes in 1st sampling 2nd sampling 1st sampling Few examples jet jet electron electron Complete list of calo shape cuts used in e-ID:  Complete list of calo shape cuts used in e-ID Leakage into the hadronic calorimeter in =0.20.2 Total leakage (typically only done at LVL1) Leakage into the 1st sampling (lass bias from noise) Shapes in 2nd sampling Shape in eta: E(37)/ E(77) Energy weighted width in 35 cells (corrected for impact within cell) Shapes in 1st sampling Fill energies in =0.125, sum over 2 bins in phi Search for local maximum cuts Energy of 2nd maximum above fluctuation E2nd – Emin above fluctuation Energy outside core frac73= [E(3s) – E(1s)] / E(3s) Width in 3 strips (corrected for impact parameter within cell) Total width in =0.1250.2 Most powerful cuts Most powerful cut Inner Detector Reconstruction:  Inner Detector Reconstruction Inner detector Pixel SCT TRT Several offline reconstruction packages available xKalman iPatRec New tracking à la DELPHI Not validated yet Supposed to be the default one in the future Note: TR information can be used for e/ separation Cut on number of high threshold TR hits (dependent on eta) Not possible for Rome productions due to problem at simulation level iPatRec track reconstruction:  iPatRec track reconstruction Default in Rome productions Muons can ‘only’ deal with iPatRec tracks Track finding Strategy Primary vertex definition Use lightweight version of track finding Simple fit to 2-point candidates (pT>3GeV) Z-coordinates stored for high quality tracks Primary track finding 2 space points + ‘loose’ transverse vertex or 3 space points from different layers Picks up all tracks from the vertex incl. those a few mm off (e.g. from B-decays) and track segments before 2ndary interactions If possible associate TRT by extrapolation + histogramming algorithm for tracks confirmed in outer SCT layer Secondary track finding 3 SCT points with very loose vertex requirement. Track confirmed if TRT associated Attempt back extrapolation towards vertex iPatRec track reconstruction:  iPatRec track reconstruction final track fit Includes material effects energy loss correction via scattering centres Less ‘aggressive’ weighting for hits shared by more than one track Single brem point may be allocated to allow for significant curvature increase in the TRT prolongation minimise gaps due to missing hits along track tracks truncated if track following fails (as in case of brem) Note: electrons with hard brem frequently found as 2 tracks Short primary one from vertex secondary one which fails back extrapolation Cluster-track match in egammaRec if E/p < 4 Position match after TrackExtrapolator:  < 0.025 and  < 0.025 (relax to 0.05 for brem tail side) xKalman track reconstruction:  xKalman track reconstruction Primary track finding starting with Pixel/SCT Original idea start from TRT and go inwards Now very similar to iPatRec Do Kalman filtering, extrapolate track to TRT Final track fit (e-fit possible (TrackModel=61 in jobO.) Track accepted if No. of prec. Hits  6 (reduces fakes) N(straw hits)/{N(straw hits)+N(empty straws crossed)} > 0.7-0.8 N(drift time hits)/N(straw hits) > 0.5-0.7 Maximise number of hits along track Cluster-track match in egammaRec if E/p < 4 Position match after TrackExtrapolator:  < 0.025 and  < 0.025 (relax to 0.05 for brem tail side) Track Quality cuts:  Track Quality cuts Used in the past for e/jet separation using xKalman No of B-layer hits  1 (ensure track comes from vertex) No of pixel+SCT hits  7 (6 for Rome layout) Transverse impact parameter  2mm For Rome data using iPatRec Transverse impact parameter |A0| 2mm All other cuts can’t be applied as tracks might be truncated Though, kind of works w/o pile-up added Track-Cluster Match:  Track-Cluster Match Match in E/p Match in  and  Extrapolate tracks into calo Info not in AOD! Cuts eta dependent Average shower depth not well parametrised in athena now Work in progress Matching in eta/phi should improve soon xKalman iPatRec xKalman iPatRec xKalman iPatRec Brem tail Jet Rejection (from DC1 data):  Jet Rejection (from DC1 data) Rejection for e25i at L= 2·1033 cm-2s-1 LVL1 looks at ET, EM and hadronic isolation using trigger towers (=0.10.1) In principle add LVL2 which works very similar as offline However, there were problem in LVL2 tracking at time of DC1 Apply ET>22 GeV in offline (to be efficiently selecting electrons with ET>25GeV) Tuning done in various eta bins to get flat response <0.8, 0.8<<1.37, 1.52<<1.8, 1.8<<2.0,<2.47 Nothing special done in barrel/EC crack besides not applying cuts in 1st sampling as there are no strips (would need special study esp. for energy reconstruction Jet rejection normalised to the number of jets reconstructed in ATLFAST Advantage: rejections can be easily ported to ATLFAST A bit more realistic due to initial/final state radiation Jet rejection at L= 2·1033 cm-2s-1 :  Jet rejection at L= 2·1033 cm-2s-1 the 29 remaining events : 21 originating from early conversion Even with conversion reconstruction package probably difficult to find these conversions (at least 1 hit in b-layer!) This tuning is implemented in athena (isem flag) Jet rejection at L= 1034 cm-2s-1:  Jet rejection at L= 1034 cm-2s-1 Optimised for electrons with ET>30GeV Again reflects inclusive single electron trigger e30i ET(EF calo) > 27 GeV 91 events remaining in final event sample 49 from conversions Possible improvements: better Brem recovery:  Possible improvements: better Brem recovery Better brem-recovery  better E/p measurement DC1, Single electrons, ET=20GeV, no pile-up, xKalman TDR data, single e, ET=30GeV, high lumi pileup in calo only Noone looked at it for a long time now… due to additional material compared to physTDR gain might be small Nice brem recovery… Material distribution changes:  Material distribution changes Major change since physTDR is insertable pixel layout Barrel: ~35% more conversions (= brem) before 3rd pixel layer ~30% more conversions before 1st SCT layer End-cap ~2 times more material in EC (2.0 <<2.5) reduced radial pixel lever arm: precision 4-6 times worse Some more material will certainly be present in ‘real’ detector… with all little bolts, supports, cables….  0 Rome-final vs. physTDR physTDR Rome-final / DC1 Possible improvements:  Possible improvements Early conversion reconstruction Has been neglected as well for a long time Problem here are fake conversions esp. in presence of pile-up Conversions near the beam-pipe are hard to reconstruct in inner detector, so probably one can’t do very much Work on conversion reconstruction about to re-start seriously Phone meeting tomorrow afternoon In addition Use beam spot to “recover” pixel lever arm Depends on physics Some correction for radiation in early layers to give correct mean of E/p for calibration Use isolation around the cluster Depends on physics, difficult in the presence of pile-up Combine energy measurement in calo and tracker for non-hard brem electrons Due to material combining tracker+calo energy measurement per definition probably very difficult Other methods to do jet rejection:  Other methods to do jet rejection Likelihoods, neural network, decision tree However, not really suited for day-1 as detailed understanding of the detector performance is needed Need to ensure works well together with the trigger selection Have to result in flat distribution in eta See talk at Rome by K. Benslama for details Don’t do into details here as no ‘robust’ results available yet What needs to be done for Rome data:  What needs to be done for Rome data New tuning for electron identification which are flat in eta Preferably include at least LVL1 trigger Unfortunately, not all cuts are available in AOD’s… Crack region should be better understood Currently 1st try Tunings should be deduced from Zee events which can be done on real data =86% =82% What needs to be done to be more ‘realistic’:  What needs to be done to be more ‘realistic’ Extract efficiencies from real data using Zee (à la CDF) similar as done in trigger (Gomes-Diaz) Use Zee events with at least one electron triggered by e25i, use other electron for the estimate Reconstruct Z-peak using e25i events and 2e25i events Number of Z events obtained from fitting procedure for e25i given by With from true Z event and from fake ones For 2e25i: Thus: Needs to be done as a function of  and pT and luminosity Some trigger related issues:  Some trigger related issues Efficiencies after each step (LVL1, LVL2, EF, offline) need to be extracted and errors understood. Fold out effects from overlaps of different trigger items e.g. e25i, 2e15i Setup new trigger menu items, e.g. e60 with 90% electron efficiencies Influence on luminosity:  Influence on luminosity Use cuts optimised using single electrons without pileup Use same cuts for electrons with pileup (L= 1033 cm-2s-1) added Try to do tuning using cuts which are less sensitive, study different lumi scenarios to find out when tunings should be changed and study effect on systematics Tuning without pileup Applied to pile-up sample Loss partly due to shower shapes, partly due to ‘missing’ tracks in sample with pile-up added Total loss: ~4% Some strange dips… InterCalibration in-situ using Zee:  InterCalibration in-situ using Zee Intercalibration needed to make uniform the response of different regions of the calorimeter Residual non-uniformities from mechanical deformations, temperature gradient… Possible with Zee (TDR) Several Methods have been proposed ATL-LARG-2004-08 (FD) 0.3% Uniformity was found in DC1 for the Barrel Method for independent phi/eta intercalibration (NK,MB) Phi with min-bias, We and +Jet Eta with Zee 0.2% Uniformity is found with full sim Caution on material effects: ATL-LARG-2004-016 (SP) Monitor calorimeter linearity with Zee Use reference Zee distribution which includes resolution effects (NK,MB) But what about higher energies… N.Kerschen, M.Boonekamp, F.Djama

Add a comment

Related presentations

Related pages

Electronic Arts Deutschland – Spiele von EA für PC, PS4 ...

We exist to inspire the world through Play. Electronic Arts is a leading publisher of games on Console, PC and Mobile.
Read more

Conrad | Der Online-Shop für Elektronik & Technik

Conrad Electronic. Sie finden uns auch auf. Facebook; Twitter; Google+; Youtube; Rss; Community; Mobile Website von Conrad.de
Read more

JZ Electronic

Hier sollte eine Beschreibung angezeigt werden, diese Seite lässt dies jedoch nicht zu.
Read more

Pollin Electronic - Shop für Elektronik, Technik und ...

Sie befinden sich auf der Seite "Pollin Electronic - Shop für Elektronik, Technik und Sonderposten". Was möchten Sie hier tun:
Read more

M.K. Electronic

M.K. Computer Electronic Ersatzteile ... diese Seite verwendet Cookies, um Ihnen das bestmögliche Shopping-Erlebnis zu bieten.
Read more


ELECTRONIC COMPANY GmbH & Co. KG, Ernst-Boehme-Str.17B, 38112 Braunschweig. Alle Rechte vorbehalten. Telefon: +49 531 23150-0 • Fax: +49 531 23150-50 • ...
Read more

Internet Seite: -www.vs-electronic.de

- www.vs-electronic.de -
Read more

DSL electronic ® GmbH - Ihr Power Solution Provider von ...

DSL electronic GmbH - Ihr Power Solution Provider von Elektronikgeräten für die Hutschiene
Read more

Aromen, Basen und mehr für die elektrische Zigarette

Bei uns steht der Kunde im Vordergrund. Wir wollen, dass Ihr als Kunden zufrieden mit uns seid und sind immer bestrebt unser Angebot zu optimieren.
Read more

Ei Electronics Rauchwarnmelder vom Testsieger ...

ei electronics Rauchwarnmelder vom Testsieger Hitzewarnmelder Kohlenmonoxidwarnmelder schützen zuverlässig.
Read more