590 likes | 711 Views
IC22 extraterrestrial cascades search unblinding proposal Joanna Kiryluk, LBNL 05/06/2009. Filtering && Data and Monte Carlo comparisons Extraterrestrial Flux Sensitivities Summary + 4. Backup slides. 1. Event Selection (shown previously). Cascade Pole Filter:
E N D
IC22 extraterrestrial cascades search • unblinding proposal • Joanna Kiryluk, LBNL • 05/06/2009 Filtering && Data and Monte Carlo comparisons Extraterrestrial Flux Sensitivities Summary + 4. Backup slides 1
Event Selection (shown previously) • Cascade Pole Filter: • LineFit Velocity < 0.25 and Tensor of Inertia Evalratio > 0.109 • Level3: Zenith>1.27 && (TrackLlh logl - CscdLlhVertex negLlh) > -16.2 • Level4: Contained events filter: Events must start inside a fiducial volume • Level5+6 Cleaning tails …. Level2 Data Burn 2 Level 2-6 Cuts description && more details in backup slides pages: 25-45
Background reduction vs cut level - Summary Bg/Signal trigger L2 L4 L6 1 Cut Level After Level6: N(Bg)/N(Signal) ~ 103 N(Signal) ~ 160 events in 240 days (E-2) 3
Cascade Filter levels comparison: Data vs MC Level 2 Level4 Level6 NChannels Excess in data Level>=4: running out of bg MC statistics 4
Cascade Filter levels comparison: Data vs MC Level 2 Level4 Level6 COGX Cascade Pole Filter Rates [Hz] COGY Shapes (COGX and COGY) ~ consistent 5
Cascade Filter : Data vs MC (1) Cascade Filter levels comparison: Data vs MC Level 2 Level4 Level6 COGZ Data/MC Rate problem: above AND below dust layer 6
Data/Corsika Rates Ratio vs cut level 1 Trigger Cascade PoleFilter Cut Level After Level6: Factor of ~3 discrepancy between Corsika and experimental rates (bg rate ~10-2 Hz) 7
Level7: Background reduction using hit topology: Fill-ratio (Doug Rutledge’s module) • distribution of distance between hit DOM and reconstructed vertex (for each event) MeanDistance x CscdLlhVertex Mean • fill-ratio= Nr of Hit Doms / Nr of All Doms within a sphere Center= CscdLlhVertex Example: Fill-ratio=2/10 = 0.2 R=2*MeanDistance x 8 Bad DOMs excluded
Fill-Ratio vs SphereRadius (Level7): Data vs MC NuE Level7 cut Above dust layer Corsika Data 10% NuE Level7 cut Below dust layer Corsika Data 10% Agreement between data and bg MC in shape above dust layer, Different shape in data and corsika below dust layer: Depth dependent cuts based on Data! 9
Background reduction vs cut level - Summary Bg/Signal 1 • No data @Level 7 • 2 corsika events@Level7 Cut Level 10
After Level7 && NCh>50 : • 56 signal events For 240 days • 2 events corsika (corresponds to 10+2.8=13 events in 240 days) • No data Example of surviving corsika event 11
Energy Reconstruction: final cut (Level8) CscdLlh energy reconstruction settings: S. Lafebre && B. Fox (PSU) NuE Bottom Top Log(RecoEn) Log(TrueEn) - Log(RecoEn) Two distributions: depth dependence not build-in to pandel phit-nohit (uses Bulk ice). Thus cannot cut on RecoEnergy without making a depth dependent correction ~ +/-0.4 12
Before RecoEnergy Correction After RecoEnergy Correction Above dust layer NuE RecoEn NuE TrueEn Below dust layer 13
MC 182 TeV e Level8 cut: Log(RecoEn_corrected)>4.2 - not optimized, due to limited bg MC statistics After final (Level8) cut, 0 events from the burn sample and corsika remain Signal e (True)Energy distribution after final (Level8) cuts Below dust layer: Above dust layer: Mean=5.26 Mean=5.23 Log(TrueEnergy) Log(TrueEnergy) • Uniform energy range for top and bottom parts of the detector 14
Position and Energy resolutions After Level8 cut on log(E_reco) 15
Energy Resolution (Level8) Below dust: Rms=0.31 Log(TrueEn) - Log(RecoEn) Log(TrueEn) - Log(RecoEn) Log(TrueEn) Above dust: Rms=0.26 Log(TrueEn) - Log(RecoEn) 16 Log(TrueEn) Log(TrueEn) -Log(RecoEn)
Position resolutions at L8 (NuE MC) for CscdLlhVertex TrueX-RecoX TrueY-RecoY TrueZ-RecoZ 17
Expected event upper limit And flux limit after unblinding For unblinding we propose to use all the cuts ( Level <=8 ) 18
Event upper limit 240 days of livetime and CL=90% Conservative limit! [ Used e.g. in HEP experiments to set limits on rare decays branching ratios ] <mu90> for bg=10 <mu90> for bg=0.0 • If much better bg MC statistics were available then we could estimate bg precisely and do bg subtraction to get a better limit (Fieldman-Cousins) • Since we don’t have enough MC, we can assume all observed events (after unblinding) are signal and set an upper limit for the mean of Poisson variable given N_observed events in the absence of background [1-CL=Sum_0^Nobs Px(n) ] 19
MRF to estimate flux limit expectations after unblinding <mu90>/nsignal for bg=10 <mu90>/nsignal for bg=0.0 • E2xlimit =MRFx10-6 [GeV s-1 sr-1 cm-2] • E2xlimit range: 7 x (10-8 - 10-7)[GeV s-1 sr-1 cm-2] (conservative limit ) • assuming n_observed events < 20 after unblinding • range depends on n_observed after unblinding • If much better MC statistics were available then we could do bg subtraction to get a better limit - but we don’t. Use limit assuming no bg. • “problem” if significantly more events observed after unblinding 20
Summary Data and MC rate discrepancy: increases with cut level. Shapes agree better than rates, bigger discrepancy below dust level (e.g. fill-ratio) Final cuts based on data (burn sample), but also on (a few) bg MC events with long-er effective lifetime than data (insufficient bg Monte Carlo statistics) Will use a conservative method to estimate flux limit assuming no background contribution (no precise estimate of background contribution thus no background subtraction method) 3. Final energy range: ~20 TeV to ~1PeV (Mean ~ 160 TeV) Expected results after unblinding: E2xlimitrange: 7 x (10-8 - 10-7) [GeV s-1 sr-1 cm-2] if n_observed < 20 after unblinding [ Flux limit will depend on number of observed events after all cuts after unblinding. Problem if “large” number of events observed] ICRC writeup (analysis description) available on the wiki page: http://wiki.icecube.wisc.edu/index.php/Papers_for_ICRC-2009 Ready to unblind 21
Data and MC Samples Data - burn sample (24 days) 2. Monte Carlo samples: a) Signal: electron neutrino: - dataset 1736 and 1739 NuGen E-2 spectrum, log10(E)= 1.6 - 10.0 Total = 3.5 M IC22 triggered events b) Background MC - Corsika (single muon) dataset 1540 (unweighted) dataset 1541 (weighted) Total= 280 M IC22 triggered events - Corsika (coincident muon) dataset 1567 (unweighted) - Corsika (tri-coincident muon) dataset 1569 (unweighted) much smaller samples: 1.8M IC22 triggered events / set 100k files generated Huge effort, production @multiple sites
Trigger and PoleCascadeFilter rates Experimental rates taken from the Monitoring page Rate [Hz] • Trigger rate: temperature variation Data: 515 Hz to 615 Hz MC: 565 Hz Data/MC Agreement • CascadeFilter rate: Data: 18 Hz to 22 Hz MC: 14 Hz Data/MC ~ 1.3 -1.6 • CascadeFilter/Trigger rate ratio: Data: 0.032-0.036 (stable) MC: 0.025 InIce SMT trigger 07/01/07 03/16/08 Rate [Hz] PoleCascade filter Time [Mjd]
Cascade Pole Filter: data vs MC P. Toale && M. D’Agostino filter proposal: LineFit Velocity < 0.25 and Tensor of Inertia Evalratio > 0.109 LineFit velocity ToI evalratio L2/Trigger Bg reduction = 0.025 Signal fraction = 0.71
Cascade Filter (Level3 FINAL): Data vs MC • Common cut for atmospheric and extraterrestrial analyse Module CscdL2Filter (Michelangelo D’Agostino UCB) Zenith>1.27 && (TrackLlh logl - CscdLlhVertex negLlh) > -16.2 • This is a loose cut for extraterrestrial analysis
Level3 optimalization for Extraterrestrial analysis Later it was decided to use looser cuts And keep the same i3 files for atmospheric And extraterrestrial analyses
Zenith [rad] RllhTrack / RllhCscd Cascade Filter Level2: Data vs MC (5) • Reconstruction results from TrackLlh and CscdLlh algorithms • Corsika (single muons): Events with Zenith>1.4 rad are misreconstructed downgoing “leading” muons • Data: Excess of ‘cascade-like’ events with small values of reconstructed Zenith angle and small values of RllhTrack/RllhCscd are for COGZ at the bottom of IC22
Data (run=109831) Corsika background Signal e Zenith RllhTrack / RllhCscd Level3 Cuts: Reconstruction Results at Level2 • TrackLlh Zenith vs RllhTrack/RllhCscd
Level3 cuts optimalisation • Sqrt(N_bg) / N_Signal vs Zenith Cut in RllhRatio bins: RllhRatio= 1.025 RllhRatio= 0.9 RllhRatio= 1.0 RllhRatio= 0.8 0.8 1.4 1.7 Zenith Cut [rad] N_Signal / sqrt(N_bg) maximum for: Zenith> 1.4 rad and RllhRatio > 1.0 (proposed cuts for this analysis at level3)
Extraterrestrial e: Level3 Cuts Optimalisation 4 TeV Find the best combination of cuts on Zenith and RllhTrack/RllhCscd by minimizing sqrt(N_bg)/N_sig (using Monte Carlo only) assuming 240 days of livetime and signal flux(es) = 1.0 x 10-6(7) E-2 . Result: Level3 cut = Zenith> 1.4 rad && RllhTrack/RllhCscd > 1.0 1.0x10^6 E-2 flux N_sig(Level3) N_sig(Level2)
Event display: Level2 Cascade Filter events 2 example events from run=109831(real data) NCh= 293 Nstrings= 19 Reco: TrackLlh Zenith= 0.59 rad Rllh-track/Rllh-cscd ratio= 0.71 NCh= 57 Nstrings= 8 Reco: TrackLlh Zenith= 0.62 rad Rllh-track/Rllh-cscd ratio= 0.83
Cascade Filter Level3 proposed cut: Data vs MC After Level3 cuts Cascade Pole Filter Improvement in Rate vs COGZ (Corsika and Data)
Cascade Filter Level3 proposed cut: Data vs MC Cascade Pole Filter After Level3 cuts Improvement in #channels distribution (Corsika and Data)
Cascade Filter (Level4): contained event filter (data) ContainedFilterModule (Steve Movit PSU) • Events must start inside Fiducial volume Level2 Data Burn + 8 DOMs top layer of the detector
Cascade Filter (Level4): contained event filter (signal MC) Before (Level3) and After (Level4) contained Filter Log10(MCPrimaryEnergy) TruePosX Log10(MCPrimaryEnergy) Geometrical effect: For cascades inside fiductial volume filter efficiency is 100% Filter most efficient at lowest energies L4/L2 Bg reduction = 0.012 Signal fraction = 0.13
Extraterrestrial e: Summary of Statistics (MC) @Trigger Level , Cascade Filter Level and Analysis-Level4 Assumed: Livetime= 240 days and signal flux = 1.0 x 10-6 E-2
T1 T2 Cascade Filter (Level5): Data vs MC (1) long tail Up to 60 s T1-T2[ns] NChannel =152 T1-T2=13 s Cut=5 s Run=109930 3-muon event First hit from the earliest muon starts in the middle of the detector, thus passing contained filter cut
Cascade Filter (Level5): Data vs MC (2) • Zenith>1.2rad downgoing muons • 32-iteration track reconstruction • Zenith>1.2rad Bg: Misreconstructed downgoing muon • Nchannels NCh>=20 (loose cut) L5/L4 Bg reduction = 0.07 Signal fraction = 0.69 running out of MC statistics
Level 5b) downgoing muons Zenith angle cut 32-iteration track reconstruction • Zenith>1.2rad for (Single) Corsika: • Misreconstructed downgoing muons • (96% of L5a NuE events survive)
Examples of ‘Cascade’ like events with highest Nchannel multiplicity at Level5
Cascade Filter (Level6): Data vs MC (1) • Reconstructed Cascade vertex and COGX (COGY) less than 60 m apart: X Y
Cascade Filter (Level6): Data vs MC (2) • Time(FirstTrack)-Track(Secondtrack) > -1000ns • (Double muon reconstruction) • Reduced Llh Track/Cascade > 0.95 L6/L5 Bg reduction = 0.37 Signal fraction = 0.98 CscdReducedLlh=(negLlh/(nHits + unHitContribution - freeParams));
ToI evalratio LineFit velocity COGZ COGX COGY Variables not usable for further cuts After Level7&&NCh>50: tightening the cuts • 56 signal events For 240 days • 2 events corsika (corresponds to 10+2.8 evts 240d ) • No data
After Level7&&NCh>50: tightening the cuts After Level7 - tightening the cuts NStrings NCh • 56 signal events For 240 days • 2 events corsika (corresponds to 10+2.8 evts 240d ) • No data Llh Diff EventLength Variables not usable for futher cuts
Extraterrestrial e: Summary of Statistics (MC)@ L4,L5 and L6 Assumed: Livetime= 240 days and signal flux = 1.0 x 10-6 E-2 MC underestimate rates: Factor of 2 (3) at L4 (6) L7&&L8 Insufficient MC Large uncertainties
‘Baloon’ events (at Level2) • small fraction observed mostly at the top part • of the detector (not x-y symmetric!) • - removed by fiducial volume cuts
Cascade Filter Level2: Total charge per Dom vs depth Shown at the Spring 2008 Collaboration meeting String 73 String 73 Dom# 6 Dom# 7 Depth dependence! <Q> [p.e.] Dom# Q [p.e.] String 39 String 39 Dom# 6 Dom# 7 <Q> [p.e.] Q [p.e.] Dom#