intercalibration of the cms electromagnetic calorimeter using neutral pion decays 1 m. gataullin...
TRANSCRIPT
Intercalibration of the CMS Electromagnetic Calorimeter Using Neutral Pion Decays
1
M. Gataullin (California Institute of Technology)M. Gataullin (California Institute of Technology)on behalf of the CMS ECAL Groupon behalf of the CMS ECAL Group
PRD08: 11th Topical Seminar On Innovative Particle and PRD08: 11th Topical Seminar On Innovative Particle and Radiation Detectors, 1-4 Oct 2008, Siena, TuscanyRadiation Detectors, 1-4 Oct 2008, Siena, Tuscany
2
CMS ECAL: 76K Crystals, 90 Tons
SSSSSshhh
Only Barrel considered: 61,200 crystals – mass 67.4 tons
170 φ-rings of 360 crystals, each ~ 25 x 25 x 230 mm3 (25.8 X0)
Test beams: energy resolution of <0.5% (~100 GeV electrons)
Calibration goal: achieve and maintain it in situ at the LHC
Barrel Barrel CrystalsCrystals
EndCaEndCapp
|η|<1.48
See Paolo Meridiani’s talk later in this session!
3
Main Purpose of the Calibration: Higgs Hunting
Crystals Pulse Amplitudes(+clustering algorithm)
Particle EnergyCalibration
• Achieving a precise in situ crystal-by-crystal calibration of the CMS ECAL
will be crucial for the Hγγ search (discovery channel for M < 140 GeV).• Design calibration precision: ~0.5%; achieved in various test beam studies.
4
π0 Calibration Concept
Level 1 trigger rate dominated by QCD: several π0‘s/event Useful π0γγ decays selected online from such events Main advantage: high π0 rate (nominal L1 rate is 100kHz !) “Design” calibration precision better than 0.5%
Achieving it would be crucial for the Hγγ detection Studies performed with about four million fully simulated QCD events. Results given for the scenario of L=2x1033cm-2s-1 and L1 rate of 10 kHz.
Data after L1 Trigger Online Farm 0 Calibration
>10 kHz~1 kHz
π0γγ Selection
5
Based on local, ECAL variables — suitable for online filter farm. Kinematics: PT (γ) >1 GeV, PT (pair) > 3.5 GeV and η < 1.4 (barrel).
Photon shower-shape cuts: S9/S25 > 0.9 and S4/S9 > 0.9, where
the sums Si are defined with 2x2, 3x3, and 5x5 crystal matrices. Isolation cut optimized to remove pairs with converted photons: Other PT in ΔR < 0.2 and Δη < 0.05 should be below 60% of PT (pair).
η
φ
Δη
γγ
ΔR
π0γγ Selection Results
6
After selection, After selection, ππ00 yield is about 0.07 per event accepted by L1 triggers. yield is about 0.07 per event accepted by L1 triggers. Signal/Background is 1.9Signal/Background is 1.9 ± 0.1, as calculated in , as calculated in ±2σ window.
The selected sample of 300,000was then used for the calibration exercises.
The π0 yield can be translated into the rate of useful decays.
Assuming a L1 rate of 12.5 kHz:
€
Rtotal( 1)L =0.87 ±0.03 kHz
Correction for Gaps and Noise Suppression
7
Corrections were derived both eta and phi directions. The dots represent the values before correction, red line – after.
Slightly bigger gaps between baskets/supermodules lead to −1.3-1.5% shifts. Selective readout: a -0.4% shift with a period of 5 crystals.
“L3” Calibration Algorithm
8
where N is the iteration step number and wi is the fraction of shower energy
in this crystal for the ith shower containing this crystal in the 3x3 matrix.
Only pairs in the peak (±2σ window) are used. Both photon energy and direction reconstructed using crystal level information (as in the selection). After each iteration events are re-selected with new constants. Calibration precision defined as R.M.S. of the products of the final and initial mis-calibration constant. Limited number of events: entire barrel folded onto a 10x10 matrix.
“Fit” Calibration Algorithm
9
For each crystal, a histogram is filled with invariant masses of pairs for which this crystal is
central (highest energy) for one of the two photons in the pair. The distributions obtained are then fitted to a gaussian+bkgd.; several iterations are required. Works because 70% of shower energy is in the central crystal. Performs slightly better than the L3 algorithm since the background shape is determined from the fit.
Performance of the Calibration Algorithms
10
Several other algorithms were investigated and found to converge to about the same final calibration precision. Final precision also does not depend on the initial mis-calibration.
4% miscalibration
no miscalibration
Calibration Results
11
The calibration precision is fitted to
€
σC
C=
a2
N
+ b2 , where N is the number of 0 per crystal
At the present level of statistical precision, we see no significant limits to improving calibration accuracy with increasing the calibration sample. Using the full sample of
3000 π0‘s/crystal, a calibration precision of 0.5% was obtained.
Studies performed with four million fully simulated events. Results given for the scenario of L=2x1033cm-2s-1 and L1 rate of 12.5 kHz. After selection, the π0 yield is about 0.07 per event accepted by L1 triggers. Using the results of the calibration exercises, the π0 rate is translated into time needed to achieve a 1%(0.5%) precision: 10 to 35 (25 to 100) hours of continuous running needed to calibrate 95% of the barrel to these levels of accuracy. Additional remarks:1) Available data throughput and CPU on the on-line filter farm will be limited. We are able to stay within the imposed constraints. 2) Situation is a bit more complex at the startup: at 10 TeV (and even
at 900 GeV) the π0 yield is lower but but still quite useful 3) The online filter for the endcaps is
also being developed.
Projected Calibration Performance
12
13
Calibration Studies in Test Beams at CERN
π0 decays produced through: π-+Al π0+X (11/2006)
Three different π- beam energies: 9, 20, and 50 GeV
Consider only 9x8 crystal matrix: about 140 π0 decays/crystal
14
First Resonance Observed by CMS
Clear improvement over the uncalibrated peak (L3 algorithm). For a precise estimate of the calibration precision: use the 50 GeV electron test beam data.
π0γγ from upstream scintillators
15
Calibration Precision with 50 GeV Electrons
For each crystal, electron energy spectra were fitted to a Gaussian.Distributions of the obtained peak positions for 9x8 crystal matrix:
Precision: 1.0±0.1% with 0.9±0.1% expected. Calibrationwith ~5 GeV photons works well for higher-energy showers!
16
Summary and Outlook
Crystal-by-crystal intercalibration to 1% should be Crystal-by-crystal intercalibration to 1% should be possible after a few days at L=2x10possible after a few days at L=2x103333cmcm-2-2ss-1-1
Optimistic outlook for achieving and maintaining aOptimistic outlook for achieving and maintaining a ~0.5% precision. Many months of work on understanding ~0.5% precision. Many months of work on understanding the ECAL performance and non-uniformity at lower the ECAL performance and non-uniformity at lower energies (work of ~15 physicists from 4 teams). energies (work of ~15 physicists from 4 teams). Test beam study demonstrated a 1% calibration precision Test beam study demonstrated a 1% calibration precision with ~5 GeV photons: successfully used to reconstruct with ~5 GeV photons: successfully used to reconstruct 50 GeV electrons, without noticeable systematic effects.50 GeV electrons, without noticeable systematic effects. Currently a lot of work is being done on finalizing filter
farm tools for collecting π0γγ in situ at the LHC. Calibration of the endcaps is also under development.
For more information on ECAL see Paolo Meridiani’s talk later in this session!The overall ECAL calibration strategy will be presented in the S. Argiro’s poster.A very important aspect of the CMS ECAL calibration&monitoring (laser monitoring system) will be presented tomorrow afternoon by A. Bornheim.
Using L1 Trigger Objects as Seeds
π0 candidates selected in regions of 20x20 crystals, containing L1 trigger electromagnetic candidates. The same selection approach with two additional cuts Nclus<4 and (Etot – E)/Etot< 0.35. A more realistic approach for the online filter farm environment. The π0 rate and S/B found to be comparable to the results obtained by selecting π0 candidates in the entire barrel. Assuming a L1 rate of 12.5 kHz:
18
€
Rtotal( 1)L =0.49 ±0.02 kHz < /S B > = 1.6 ±0.1
Selection Results
19
After selection, high rapidity regions suffer both in the event rate andSignal/Background (background rate is almost constant with η).