Common Point source analysis from IC22 & Antares 2007/2008 data sets J. Brunner DESY/CPPM Motivation 4 skymap of TeV neutrinos Improved limits for selected sources (or selected declination bands) Reproduce individual results using one coherent method Ensure that Antares/iceCube compare “apples to apples”

Region of sky observable by neutrino telescopes Mkn 501 Mkn 421 CRAB SS433 Mkn 501 RX J GX339-4 SS433 CRAB VELA Galactic Centre IceCube (South Pole) ANTARES (43° North) <25% exposure not visible Intriguing complementarity of sky coverage

Start with most basic one Start with most basic one First step: Fixed point source analysis First step: Fixed point source analysis One free parameter to fitOne free parameter to fit number of signal event on given sky locationnumber of signal event on given sky location This presentationThis presentation Later extension Later extension Full sky searchFull sky search 3 parameters to fit: location & signal strength 3 parameters to fit: location & signal strength Addition of energy estimatorAddition of energy estimator Fit also spectral index (2 or 4 parameters to fit) Fit also spectral index (2 or 4 parameters to fit) Untriggered flare search DESY)Untriggered flare search DESY) Which type of analysis to choose ?

Analysis Method ? Unbinned search, likelihood Ratios: used in both collaborations for most recent searches ANTARES: S+B likelihood ANTARES: S+B likelihood signal PDF Signal events Poisson mean PSF Position in sky Background rate Total event Derived from binned method with bin size which becomes zero (consider only 0 or 1 event per bin)

Analysis Method ? Unbinned search, likelihood Ratios ANTARES: S+B likelihood ANTARES: S+B likelihood signal PDF Signal events Poisson mean PSF Position in sky Background rate Total event Full sky search:3 parameters fitted (,,) Predefined list: only  fitted So far no energy estimator used : E -2 flux fitted (fixed spectral index) Test statistics: ratio of signal/ background likelihood

Analysis Method ? From IC40 paper Similar concept, some subtle differences Test statistics B i in IceCube pure PDF, whereas in Antares denotes event number N = total event count constant in IceCube Bg = total background constant in Antares IceCube: S+B likelihood IceCube: S+B likelihood

Analysis Method Differences IceCube: energy proxy used & spectral index fittedIceCube: energy proxy used & spectral index fitted Remedy : Integrate over E -2 flux (T.M.:40% effect) Remedy : Integrate over E -2 flux (T.M.:40% effect) IceCube: Event-by-event angular error used in PSFIceCube: Event-by-event angular error used in PSF Remedy: Use average PSF (tiny effect) Remedy: Use average PSF (tiny effect) Different Likelihood definitionDifferent Likelihood definition Try both to see the effect Try both to see the effect Problem: works fine with Antares LH Problem: works fine with Antares LH No way to get LH with maximum for IceCube No way to get LH with maximum for IceCube N>>n s therefore L maximal if ns maximal ???? Rest of the talk: use Antares convention

Combining Samples (no energy used) Add sum of j samples Add sum of j samples Each event knows from which sample it comes (indices ij) Each event knows from which sample it comes (indices ij) Each sample has its own background PDF B j Each sample has its own background PDF B j Each sample has its own PSF ß j Each sample has its own PSF ß j But only one  sig can be fitted But only one  sig can be fitted Relative contributions of samples to signal controlled by r j for each  Relative contributions of samples to signal controlled by r j for each 

Data taking periods & analysis samples ANTARES: analysis on 2007/2008 data ANTARES: analysis on 2007/2008 data IceCube: IC22, IC22-PeV, IC40 IceCube: IC22, IC22-PeV, IC40 Good overlap Good overlap IC40IC59IC days 5114 ev S 1877 ev N [-50,+85] 375 days ev S ev N ANT-5ANT-10-12ANT days 740 ev S 60% 151 days 1300 ev S 60% ? A. Heijboer J. Dumm R. Lauer

PointSpreadFunction Cross check: PSF symmetric Cross check: PSF symmetric 1-dim parameterization sufficient 1-dim parameterization sufficient Example: Antares Example: Antares

Point Spread Function Antares Integrate over weak declination dependence Integrate over weak declination dependence Defined in log of angular error Defined in log of angular error PSF defined over full angular range PSF defined over full angular range º º 2008

PSF IC22 Integrate over energy for E -2 flux Integrate over energy for E -2 flux Energy integration UHE analysis

PSF IC22 Energy integration Cumulative distribution Median 1.5º PSF defined up to 5 degrees PSF defined up to 5 degrees More than 90% of events included within 5º More than 90% of events included within 5º Defines natural range for LH definition and minimization Defines natural range for LH definition and minimization  apply this at all subsequent steps  apply this at all subsequent steps

Background PDF Antares: generation of pseudo- experiments from spline fits to data Antares: generation of pseudo- experiments from spline fits to data No use of scrambled data No use of scrambled data events 740 events

Background PDF IC22 Event density in bins of 2º radius Event density in bins of 2º radius Convert to make coherent to Antares Convert to make coherent to Antares UHE-analysis

Background PDF IC22 Convert into event density per angular bin Convert into event density per angular bin Select downward part of Robert Lauer sample Select downward part of Robert Lauer sample UHE-analysis Use only downward events Avoid double counting Events per angular bin 5114 events 788 events Bump Dip Visible in sky maps:

ANTARES signal events ANTARES signal events r j calculation and Nevent  flux conversion Benchmark flux Benchmark flux 2007: 5-lines 0.9 events max 2008: 10,12-lines 2.3 events max declination ANTARES 2007 ANTARES 2008

IC22 Signal events IC22 Signal events r j calculation and Nevent  flux conversion Benchmark flux Benchmark flux Effective area in declination bands UHE Analysis

IC22 Signal events IC22 Signal events r j calculation and Nevent  flux conversion Benchmark flux Benchmark flux Effective area in declination bands After energy integration 50 events max declination Coarse declination binning: interpolate & rebin

Signal events Unified in bins of 10 degrees Unified in bins of 10 degrees Event per declination per reference flux Event per declination per reference flux Used for signal simulation and r j calculation Used for signal simulation and r j calculation ANT07 ANT08 IC22-UHE IC22 ALL

Signal events Fraction of events per declination band Fraction of events per declination band Northern hemisphere: IC22 contributes more than 95% Northern hemisphere: IC22 contributes more than 95% DEC<-50º: only ANTARES DEC<-50º: only ANTARES -50º -0º: ok -50º -0º: ok ANT07 ANT08 IC22-UHE IC22 Declination range where both experiments contribute more than 10% Chance to get events from both samples close to source

Code development Signal histos Background histos PSF histos Test experiments Fit within 5º from source Filenames NEXP Nsig Ra,Dec Command line Small root files Root Tree TS,LH(S+B),LH(B),sig SensitivityDiscovery powerFlux limits Root executable No graphics Runs in batch mode Small Root Macros Graphics output Interactive

Code development Everything is based on Root Everything is based on Root No dependence on heavy software packages like IceTray, SeaTray No dependence on heavy software packages like IceTray, SeaTray No dependence on custom formats like i3 No dependence on custom formats like i3 Test experiment drawing and LH fit coupled Test experiment drawing and LH fit coupled Restrict to 5º area around source for speed reasons Restrict to 5º area around source for speed reasons Heaviest part standalone executable Heaviest part standalone executable Mass production on batch farm at CC-LyonMass production on batch farm at CC-Lyon 18 declinations (-85º to 85º) in steps of 10º18 declinations (-85º to 85º) in steps of 10º Fixed (random) rectascensionFixed (random) rectascension Signal events from 0-29 : Total 540 jobs per roundSignal events from 0-29 : Total 540 jobs per round 10 7 TE for BG-only case, 10 6 otherwise10 7 TE for BG-only case, 10 6 otherwise

Example sky maps Point source with 20 events included Point source with 20 events included Rectascension Declinationion IC22 IC22-UHE ANT-2007 ANT2008

Example sky maps Point source with 20 events included Point source with 20 events included Rectascension Declinationion IC22 IC22-UHE ANT-2007 ANT2008

Example sky maps Point source with 20 events included Point source with 20 events included Dominated by IC22 events Dominated by IC22 events Rectascension Declinationion IC22 IC22-UHE ANT-2007 ANT2008

Example sky maps Point source with 20 events included Point source with 20 events included Rectascension Declinationion IC22 IC22-UHE ANT-2007 ANT2008

Example sky maps Point source with 20 events included Point source with 20 events included Rectascension Declinationion IC22 IC22-UHE ANT-2007 ANT2008

Example sky maps Point source with 20 events included Point source with 20 events included Reasonable mixture of both sets Reasonable mixture of both sets Rectascension Declinationion IC22 IC22-UHE ANT-2007 ANT2008

Examples for TestStatistics More pronounced change in TS in Southern hemisphere More pronounced change in TS in Southern hemisphere Less Background Less Background signal events DEC=-35º DEC=+35º

Examples for Fitted Nsig Antares Antares fitted values are systematic 8% lowfitted values are systematic 8% low Effect of 5º cutoff in PSF for fitEffect of 5º cutoff in PSF for fit IceCube IceCube Fitted values ok within 2%Fitted values ok within 2% Effect ignored in the following Effect ignored in the following signal events DEC=-35º DEC=+35º fitted fitted

Discovery power Cumulative TS distribution for BG-only experiments Cumulative TS distribution for BG-only experiments Exponential fit Exponential fit Retain 3, 5 TS values from single-sided Gaussian ( , ) Retain 3, 5 TS values from single-sided Gaussian ( , ) DEC=-35º DEC=+35º

Discovery power Typical quote of discovery power: 50% chance to have a 3, 5 effect Typical quote of discovery power: 50% chance to have a 3, 5 effect DEC=-35º Single source DEC=-35º 24 sources approximation DEC=+35º 24 sources approximation

Results : Discovery power Declination dependence over full sky Declination dependence over full sky Trial factor for 24 sources included in approximate way Trial factor for 24 sources included in approximate way 3 53 5 3 53 5 Event numbers Neutrino Flux

Sensitivity (average upper limit) In principle same method In principle same method Need median from BG-only TS distribution, but large peak at TS=0 Need median from BG-only TS distribution, but large peak at TS=0 Look at Nsig (Flux) for which 90% of experiments above this value Look at Nsig (Flux) for which 90% of experiments above this value Different statistical approaches used Different statistical approaches used Neyman: same as for discovery power P(S+B)Neyman: same as for discovery power P(S+B) Best limits but sometimes 0 excluded Best limits but sometimes 0 excluded Feldman&Cousins (not yet implemented)Feldman&Cousins (not yet implemented) Exact coverage without above problem Exact coverage without above problem CLs : P(S+B)/P(B)CLs : P(S+B)/P(B) overcoverage overcoverage Effect of background fluctuations largely eliminated Effect of background fluctuations largely eliminated Conservative (slightly worse limits) Conservative (slightly worse limits)

Comparison of limit setting methods DEC=+35º Neyman Median=3.8 DEC=-35º Neyman Median=2.7 DEC=+35º CLs method Median=5.8 DEC=-35º CLs method Median=3.8 DEC=-70º Antares Neyman 2.0 F&C 2.4 CLs 2.8

Limit setting methods Difference of up to 50% between different methods are observed Difference of up to 50% between different methods are observed IceCube and Antares have both adopted F&C IceCube and Antares have both adopted F&C This analysis: results for Neyman & CLs are shown (F&C just in the middle) This analysis: results for Neyman & CLs are shown (F&C just in the middle) T.M.: “Comparing Neyman versus F&C in IceCube makes a 15% effect” T.M.: “Comparing Neyman versus F&C in IceCube makes a 15% effect”

Sensitivities in comparison ANTARESIceCube Sensitivity is average of actual limit points 50% of points above, 50% below Sensitivity coincides with best limits i.e. those which are most BG-like Not entirely understood: Difference in likelihood definition Difference in treatment of TS=0 peak for bg-only test experiments

Results : Sensitivity - Nsig 90% C.L. Neyman CLs

Results : Sensitiviy - Flux Reproduce Antares & IC22 where one experiment dominates Reproduce Antares & IC22 where one experiment dominates Coherent limit for whole sky Coherent limit for whole sky 90% C.L. Neyman CLs IC22 IC22-UHE Antares

IC22 source list All sources on Northern hemisphere All sources on Northern hemisphere Less interesting for this analysis Less interesting for this analysis

Antares Source list Galactic Galactic 7 sources ok 7 sources ok Extra- Galactic Extra- Galactic 8 sources ok 8 sources ok

Next steps Short term Short term Make IceCube Likelihood workMake IceCube Likelihood work Compare both LH definitionsCompare both LH definitions Implement F&C method (if needed)Implement F&C method (if needed) Define source listDefine source list Median term Median term Implement full sky search (3-param fit)Implement full sky search (3-param fit) Request real data (“unblinding”)Request real data (“unblinding”) Write internal note (astro-ph ?)Write internal note (astro-ph ?) Longer term Longer term Use energy estimator (wait for Antares)Use energy estimator (wait for Antares) Try untriggered flare searchTry untriggered flare search careful with up-time corrections between IC & ANT careful with up-time corrections between IC & ANT Try different data sets (?)Try different data sets (?)