G. BrunoOffline week - February Comparison between test- beam data and the SPD simulations in Aliroot G. Bruno, R. Santoro Outline: strategy of the MC simulation comparison with real data conclusions
G. BrunoOffline week - February Minibus 2 Minibus 3 Test chip 425 m 50 m 256 rows 32 columns Beam test in 2003 Test detector: 300 m sensor Tracking precision: (x) = (y) 10 m Full scan of threshold and tilt-angle y x
G. BrunoOffline week - February Minibus 0 Minibus 1 Test chip 425 m 50 m 256 rows 32 columns Beam test in 2002 Test detector: 200 m sensor Tracking precision: (y) 6 m threshold and tilt-angle scan y x
G. BrunoOffline week - February Kinematics: – -, p= 120/350 GeV, gaussian beam profile ( x = y =0.2 cm) –Beam focusing tuned to reproduce the real data –1 track per event ; 50K events for each setup (threshold, tilt-angle, etc) Geometry: –Starting point: AliITSvSPD02 (setup for 2002 by J. Conrad e B. Nielsen) –Our developments (actually a minor work): setup for 2003 geometry with test-plane tilted (for both 2002 and 2003) thin sensor (200 m) with thick chips (750 m) for 2002 setup SPD response-function & simulation: –AliITSresponseSPD, AliITSsimulationSPD (i.e. the Ba/Sa model without diffusion) Strategy of the MC simulation Ba/Sa
G. BrunoOffline week - February Kinematics Geometry SPD response-function & simulation (cont.) –AliITSresponseSPD, AliITSsimulationSPDdubna (i.e. the Dubna model with the diffusion) Clustering, tracking, efficiency and precision studies are done with the same codes used for test-beam real data (see talk by D. Elia) –Immediate comparison –No bias from different algorithm Strategy of the MC simulation Dubna
G. BrunoOffline week - February Tracking precison: setup 2003 track (x) track (y) 8 m For real data: track = 10 m
G. BrunoOffline week - February Tracking precison: setup 2002 track (y) 5 m For real data: track (y)=6 m
G. BrunoOffline week - February Comparison Ba/Sa MC vs. data Setup 2003 –P c =P r =0. (no coupling) P c =P r =0.1 (suggested coupling) Real MC P=0.1 MC P=0 Coupling has to be introduced !
G. BrunoOffline week - February Comparison Ba/Sa MC vs. data Ba/Sa MC Real data One might play with P r and P c (let’s say P r =0.2 P c =0.03 ) P c =P r = but this would mask the real physics ongoin in the detector !
G. BrunoOffline week - February Comparison Ba/Sa MC vs. data Setup 2003P c =P r =0.1 (in ALICE notes) Ba/Sa MCReal data 3 2 1
G. BrunoOffline week - February Comparison Ba/Sa MC vs. data Setup 2003P c =P r =0.1 (in ALICE notes) Ba/Sa MCReal data 3 1 Even if cluster type distribution can be reproduced, it will not be related with track impact on the pixels
G. BrunoOffline week - February Comparison dubna MC vs. data Setup 2003 –P c =P r =0. (no coupling) –standard conditions for diffusion –E th = 3220 elec/holes Real MC Coupling can help with the fine details
G. BrunoOffline week - February Comparison dubna MC vs. data dubna MC Real data
G. BrunoOffline week - February Comparison dubna MC vs. data dubna MC Real data In log scale
G. BrunoOffline week - February Comparison dubna MC vs. data Setup 2003 dubna MCReal data 3 2 1
G. BrunoOffline week - February Comparison dubna MC vs. data Setup 2003 dubna MCReal data MC distribution is narrower than real data: not enough diffusion in the model ! 3 1
G. BrunoOffline week - February Comparison dubna MC vs. data dubna MCReal data Efficiency versus threshold parameters Is there a relation between DAC and MC th ?
G. BrunoOffline week - February Comparison dubna MC vs. data dubna MCReal data Efficiency versus threshold parameters gaussian fit no linearity Real data: threshold linear over the full range (see talk by Domenico) ! MC: at very hard threshold linearity is lost ! gaussian fit
G. BrunoOffline week - February Comparison dubna MC vs. data dubna MCReal data Efficiency versus threshold parameters
G. BrunoOffline week - February dubna MC This naive method can give a good estimate ! Comparison dubna MC vs. data no MC linearity
G. BrunoOffline week - February Comparison dubna MC vs. data dubna MCReal data Precision of the tracking is a bit better in the MC –it is better to compare the intrinsic resolutions
G. BrunoOffline week - February Comparison dubna MC vs. data dubna MCReal data 200 m steeper than 300 m both in data and in MC m: there is a maximum as observed in real data m: the minimum cannot be reached: one has to introduce more diffusions in the model !!!! nominal precision
G. BrunoOffline week - February Comparison dubna MC vs. data dubna MCReal data 300m Threshold (e - ) With more diffusion in the model the cl2 curve is expected to go up !
G. BrunoOffline week - February Comparison dubna MC vs. data dubna MCReal data 200m Threshold (e - ) Again, with more diffusion the cl2 curve should go up (but less than at 300 m)
G. BrunoOffline week - February Definition of cluster types VTH = 200
G. BrunoOffline week - February tilted angle 0° 300 m e-h 300 m MC Comparison dubna MC vs. data data (DAC 190) MC (3220 e - ) * For a given threshold DAC one can already get a good matching by playing only with E th
G. BrunoOffline week - February tilted angle 0° 300 m e-h 300 m MC Comparison dubna MC vs. data 0O0O
G. BrunoOffline week - February m e-h 300 m MC tilted angle 10° Comparison dubna MC vs. data
G. BrunoOffline week - February m e-h 300 m MC tilted angle 20° Comparison dubna MC vs. data
G. BrunoOffline week - February cpu consumptions in the two models hitssdigitshitssdigitshitssdigits Ba/Sa cp time real time 0:00:380:00:180:17:440:08:037:03:373:46:20 Dubna cp time real time 0:00:38 0:04:020:17:430:46:156:55:496:37:31 1K events 10K events 50K events The Ba/Sa code is much faster for small size file (the model is simpler) but both become slow when managing large files
G. BrunoOffline week - February Conclusions As it is, the Ba/Sa model is not suited for studies such as charm and beauty production (displacement of the secondary vertices) The dubna model reproduces the test beam details much better In term of cpu, dubna slower than ba/sa Test beam data suggest that more diffusion has to be introduced in the model
G. BrunoOffline week - February What next Fine-tuning Optimization of the algorithm in term of cpu
G. BrunoOffline week - February A reminder of the Ba/Sa model The energy deposited in the sensitive material during the transport (at the moment GEANT) is distributed among the pixels according to two mechanisms: Charge sharing –Energy in each pixel proportional to the track path in that pixel Capacitive coupling between adiacent pixels –P c (P r ) is the probability to fire an adiacent pixel along the column (row) –If fired, it gets the same energy E of the parent pixel –Default: P c =P r =0.1 (in Aliroot set to 0) not fired Fired, Ecoupl, E
G. BrunoOffline week - February references: R. Caliandro, R. Dinapoli, R. A. Fini, T. Virgili, Simulation of the response of a silicon pixel detector, Nucl. Instrum. Meth. A 482 (2002) R. Caliandro, R. Dinapoli, R.A. Fini and T. Virgili, A model for the simulation of the response of pixel detectors, ALICE INT R. Caliandro, R. Dinapoli, R.A. Fini and T. Virgili, Simulation of the response of the ALICE silicon pixel detectors, ALICE INT R. Barbera, R. Caliandro, B.V. Batyunya, A.G. Fedounov, R. A. Fini, B.S. Nilsen, T. Virgili, Status of the simulation for the silicon pixel detector in ALICE, ALICE-INT A reminder of the Ba/Sa model At the initial stage, noise is added to all the pixels according to a gaussian (default: sigma = 280 elec-hole pairs) From Sdigit (analog) to Digit (digital) –A pixel gives a digit if the energy is larger than a threshold E th (default: E th =2000 elec.-hole pairs) Actually the model was thought with a parametrization of the diffusion –never implemented in Aliroot
G. BrunoOffline week - February A reminder of the Dubna model Charge sharing: diffusion –The electrons/holes produced along the track are let to diffuse (T,V, ). –Slower than Ba/Sa (we will quantify later) –A better physical description: essential to match the observed (real data) improvements in the intrinsic resolution due to cluster 2,3 Capacitive coupling between adiacent pixels –the same as Ba/Sa –By default switched off: P c =P r =0.0 noise: –“electronics” = “baseline” + “noise” (i.e. const + gaussian) –Default: “electronics”=0.+0. After work by B.Nielsen and J. Conrad for merging features of the two models