Separation of Cosmic-Ray Components in Water Cherenkov Detector and use of neural networks to measure m/EM Luis Villaseñor* and H. Salazar FCFM-BUAP 5th International Workshop on Ring Imaging Cherenkov Detectors Playa del Carmen November 30 – December 5, 2004 *On Leave of Absence from IFM-UMSNH
Contents Motivation to study m/EM separation Experimental setup Data Composition of showers with known m/EM Use of neural networks Conclusions
Nm/Ne Strongly Correlated With Primary Mass, i.e. ~2 x for Fe wrt p
Look here To understand there Use low energy data to get real m and EM traces to eliminate systematics due to detector simulation Look here To understand there
Measure Charge, Amplitude,T10-50,T10-90 with good precision for three different triggers. Arbitrary muons threshold of 30mV
1.54 m diameter, 1.2 m water, one 8” PMT, tyvek 1/5 in volume of an Auger WCD
LabView based DAS
~74 pe
R shower (Q>7VEM) = 1 Hz Low Charge Peak=0.12 VEM R muon = 876 Hz R EM = 80 Hz R shower (Q>7VEM) = 1 Hz Not an Artifact due to V threshold
Stopping muon at 0.1 VEM Decay electron at 0.18 VEM Crossing muon Alarcón M. et al., NIM A 420 [1-2], 39-47 (1999).
Decay electron Stopping muon at 0.18 VEM at 0.1 VEM In this case Qpeak=0.12 VEM EM particles of ~ 10 MeV Crossing muon at 1 VEM
With PMT Glass Cherenkov signal
No PMT Glass Cherenkov signal
With PMT Glass Cherenkov signal
No PMT Glass Cherenkov signal
Separation of individual Muons and EM particles is Easy for low energy Calibration events
Stopping muon or electron Q~0.12 VEM (9 pe) T12~3ns Isolated Muon Shower Q>7 VEM (500 pe) T12>15ns
4 muons, 15ns Data trace Q=7.8 VEM 8 muons 15 ns 33 “electrons” 25 ns
Parameters for Data and Composed Events Charge (VEM) 7.9+-0.5 8.0+-0.55 7.9+-0.51 8.34+-0.4 Amplitude (V) 1.16+-0.08 1.20+-0.20 1.25+-0.20 1.34+-0.19 T10-50 (ns) 16.7+-0.9 17.5+-3.0 18.25+-3.6 18.45+-2.9 T10-90 50.8+-2.0 50.0+-4.3 52.4+-6.6 54.2+-6.9
2 or 3 classes as output (8m, 4m + 33e, 66e) Training and Clasification Results for a Kohonen Neural Network 4 features as input (Charge, Amplitude, T10-50, T1090) 8 Neurons in first layer 4 in second layer 2 or 3 classes as output (8m, 4m + 33e, 66e)
Training and Clasification Results for Two Classes 8 m 4m 33 e Data 65% 39% 68% 35% 61% 32% Class
Training and Clasification Results for Two Classes 8 m 0m 66 e Data 65% 33% 78% 35% 67% 22% Class
Training and Clasification Results for Three Classes 8 m 0 e 4 m 33 e 0m 66 e Data 56% 29% 33% 58% 21% 35% 27% 15% 0 m 23% 36% 40% Class
Conclusions Clear separation of muons, electrons, PMT interactions and showers in a single WCD Rise time 10-50% is linear with Q/V Neural Networks classify composed events of muons and electrons better than randomly Shower data is dominated by muons To do: Apply to Auger with 25 ns sampling time.