110 likes | 116 Views
This document discusses the corrections applied to improve energy resolution in the EM cluster. Topics include S shape in eta, phi offset, gap correction, longitudinal weights, and more.
E N D
G.Unal 9/05/05 EM cluster corrections • Status in 10.0.1 (default « Rome » option) • Corrections applied: • 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) • E vs eta local modulation f(Energy) • Gap correction f(eta) • Longitudinal weights (« lwc904gap ») f(eta) • Everything derived from G4 single electron samples (Scott+Stathes) • 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 cone corrections aborbed in Scale(eta) • Some corrections different for 5x5, 3x5 3x7
Performances in 10.0.1 • Some checks shown by Scott and Stathes at the last calo performance meeting • Various plots at the AOD level shown (e.g. Fabiola at the last phone meeting on Z->ee, in the barrel the average energy resolution is 1.9%, probably not as good as TDR) • Try running 10.0.1 « out of the box » on single electrons made with 9.0.4 (Rome simulation version) in/castor/cern.ch/grid/atlas/wisc/simul/single/simul.singleel_*.root • Digitize (without electronic noise) + reconstruct (Lar EM clusters) • Look at few eta and energy points. Fit gaussian part of the resolution (-1.5, 2) sigmas
Fit sampling term: eta sampling term (sqrt(GeV)) 0.0625 9.9% 0.6125 11.0% 1.0875 16.0% (+- 0.5%) 1.2625 18.2% 1.5125 70% 1.8125 14.4% 2.0125 11.8% Results similar to plots shown by Scott in Mach For eta>1 in the barrel, results are ~15-25% worse than values quoted in the TDR (for instance 12.5% at 1.1) At smaller eta, results are closer to TDR (but still slightly worse)
Switch off layer weight and look at resolution vs PS weight. (remember that 5% sampling fraction is applied at the cell level <=> weight of ~4) • Weight used in lcw904 (0.96) is ~ OK Cannot really improve with this correction procedure
Lateral leakage fluctuations seem also to play an important role in the resolution: eta=1.0875 , fitted sampling term sliding window corr. 3x7 16.0% sliding window uncorr. 3x7 16.3% naive 3x7 16.0% moving 3x7 (to max. energy) 15.7% (small improvement ?) naive 9x9 14.8% naive 11x11 14.1% all LAr cells 12.7% ?
Eta=1.0875 Eta=0.0625
Improvements ? • Separation of energy lost before PS and energy lost between PS and strips (cf Test Beam). Energy dependent weights ? • use calibration hits to compute these weights • Probably will not fully recover lateral leakage fluctuations. • Correct fluctuations in lateral leakage ? • See Stathes and Leonardo presentations in previous meetings. Use variables like shower width, shower depth ?
From Leonardo talk at last Lar week Photons at eta=0.31 Lateral containment • First studies in the simplest case: a fixed 3x5 (default for unconverted photons) clustering algorithm implemented at the hits analysis level • Correction 5% • Correction variation 2% Important correction should not be ignored!! Does this works for electron at larger eta ?
From december discussion EM cluster Corrections Correction list: (ordered following discussion with Dirk,David,Srini,Scott) • Eta 2nd sampling S shape • Phi offset • Eta 1st sampling S Shape • Out of coneshort term: merged together in form • Gap (crack Barrel-EndCap) S(A+B*PS+E1+E2+W*E3) • “Longitudinal” weights (+Overall scale) Longer term; disentangle a la TB analysis • E vs local phi mod (-> should also be extented for local HV residual effects) • E vs local eta mod • (Photon pointing also needs shower depth parameterization) • f(E,type,overall eta) (overall phi for upstream dead matter ?) and cluster size for E corrections (type=e or gamma, e+ vs e- could also be relevant in some cases) • "Old" scheme: cells weights applied before clustering for energy lost upstream/downstream (now: PS has effective samping fraction of 5% barrel, 1.667% End-Cap) • Position corrections have to be done before E local modulation • Local mod. should not change the overall scale (average=1) • Interplay between out-of-cone,gap,long.weights not trivial • Is “raw” E good enough for E dependence of position corrections ? Seems yes • Local modulation corrections can be skipped at Lv2 (and S shape 1st sampling, phi offset) • Fine details of some corrections probably correlated to identification cuts… Then, should be able to apply photon/electron specific corrections.