CMS Calorimeter

1. CMS Calorimeter CMS Calorimeter (ECAL+HCAL) - Very hermetic (>10λ in all η, no projective gap) HO-2 HO-1 HO0 HO+1 HO+2 HB- HB+ HE- HE+ HF- HF+ Tracker EE- EE+ EB- EB+ Super conducting coil Return yoke Muon chambers HB Brass Absorber (5cm) + Scintillator Tiles (3.7mm) Photo Detector (HPD) |h| 0.0 ~ 1.4 HE Brass Absorber (8cm) + Scintillator Tiles (3.7mm) Photo Detector (HPD) |h| 1.3 ~ 3.0 HO Scintillator Tile (10mm) outside of solenoid Photo Detector (HPD) |h| 0.0 ~ 1.3 HF Iron Absorber + Quartz Fibers Photo Detector (PMT) |h| 2.9 ~ 5.2

3. We observe two classes of anomalous signals in HCAL 1) Electronics noise from the Hybrid Photo Diode (HPD) and Readout BoX (RBX) used for the Hadronic Barrel (HB), Outer (HO), and EndCap (HE) calorimeters The HPD has 18 channels/device There are 4 HPDs in a RBX HV gap ~8kV 2) Cherenkov light produced by interactions in the window of the Forward Calorimeter PMTs Glass window thickness in the center is ~1mm increasing to ~6.1mm on the edges

4. HPD and RBX Noise HPD Ion Feedback (~1 HPD channel) Photoelectron induced liberation of ions from the silicon diode which accelerate across the HV gap and interact with the photocathode freeing additional photoelectrons HPD Discharge (up to 18 HPD channels) With the HPD operating at ~8kV in the CMS magnetic field, dielectric flashover from the wall can produce large signals in many channels RBX Noise (up to 72 channels) Source unknown, possibly due to external noise coupling to HV of many channels across the whole RBX CMS Preliminary For the HB and HE the total rate of HPD and RBX noise is 10-20 Hz for E>20 GeV which includes the contribution from the OR of 288 HPDs HPD and RBX noise is random and the overlap with physics is very low HPD/RBX noise produce distinct patterns in HCAL Filters have been developed making use of hit patterns, timing, pulse shape, and EM fraction

5. Signals from interactions in the HF PMT windows and in the fiber bundles were observed in the testbeam and published in: Eur. Phys.J. C53, 139-166 (2008) Long fibers: extends for the full length of HF Short fibers: start at a depth of 22cm from the front of HF Most of the HF PMT hits can be identified based on the energy sharing between the Long and Short fibers using R = (EL-ES)/(EL+ES). Filters have been developed to effectively remove anomalous signals with little impact on real energy deposits. Rates in 2009 minimum bias data (~1720 PMTS) ~6x10-3 per event identified for 900 GeV data ~8x10-3 per event identified for 2.36 TeV data Dominant sources are muons from decays in flight and hadron shower punch through More details on the filtering and characteristics of the noise has been published in: JINST 5 T03014 2004 Test beam EL (GeV) PMT Hits Fiber Bundles ES (GeV) CMS Preliminary PMT Window

8. PET Cleaning Algorithm • The topology of PMT window hits can be compared to the expected longitudinal and lateral shower profiles in HF • Polynomial Energy Threshold (PET) algorithm flags a reconstructed hit in a short fiber as a PMT hit if its energy, ES, is • i) above some energy threshold, and • ii) very large compared to the long fiber energy, EL, • in the same HF tower. The energy threshold is a polynomial function of η, ranging from about 35 GeV (at η = 3) to 50 GeV (at η = 5). If this threshold is passed, and the ratio R = (EL − ES )/(EL + ES) < 0.8, the cell is identified as a PMT hit

9. S9/S1 Cleaning Algorithm • S9/S1 algorithm, used for long fibers, identifies PMT hits by comparing the energy in a long fiber to the sum of energy in 9 of its neighbors (4 long fibers and 4 short fibers in the adjacent HF towers, plus the short fiber of the same HF tower). • The isolation variable “S9/S1” is defined as a ratio between the energy sum of the 9 neighbors, S9, and the energy of the long fiber cell under consideration, S1. • If this ratio is smaller than a threshold, defined as a function of long fiber energy, the cell is flagged as a PMT hit.

10. Timing/Pulse ShapeBased Cleaning • In addition to topological cuts, the pulse shape/timing information of the signals is used to identify PMT hits in both long and short fibers • Real signals are expected to peak in time sample 4 for all channels in HF. A charge ratio • provides good discrimination between real energy and PMT window hits. Real signals create charge deposits confined almost completely within time sample 4, resulting in a ratio Q ≈ 1. Anomalous hits have lower Q values • An HF RecHit with energy greater than 40 GeV is flagged as noise if its Q ratio satisfies the condition: • The combination of pulse shape and topological cleaning provides a set of complementary criteria to identify anomalous signals in HF

11. Filtering Performance • The probability to incorrectly flag a real energy deposit in HF with ET > 5 GeV as an anomalous hit was determined to be ~10−3. This value is considered small enough to allow the application of this cleaning by default in the standard CMS reconstruction • The noise cleaning algorithms strongly reduce the tails of the MET distribution caused by HF anomalous noise • Some events due to residual HF noise are still visible in the tail of the MET distribution. They have been visually inspected and classified as • i) double-hits in the same HF tower, • ii) multi-hits affecting several channels typically in the same φ strip, and • iii) PMT hits embedded inside a jet. • Such residual anomalous hits, not flagged by the existing algorithms and producing high MET (MET > 45 GeV), occur at a very small rate of approximately 10−4 with respect to the total rate of ~10-2 anomalous hits/event. New criteria to identify this kind of noise are currently under study