Search for Point-like Sources with the ANTARES Neutrino Telescope Francisco Salesa Greus IFIC (CSIC Universitat de València, Spain) On behalf of the ANTARES collaboration
Outline Neutrino astronomy. The ANTARES detector. Methods of point-like sources search: Binned methods. Unbinned methods. Results with 2007 data. Fixed-sources list. All sky. Summary. 2
Neutrino astronomy Astrophysical sources Binary µ-q systems AGN AGN Micro- quasars SNR GRB GRB Galactic sources Extragalactic sources Dark matter: annihilation of neutralinos in massive objects (Sun, Galactic Centre, ) Neutrino oscillations: atmospheric neutrino angular distribution. Monopoles, top-down models, etc. 3
Detection principle Neutrino Astronomy Cosmic neutrinos interact with the surrounding medium. The produced charged particles induce the Cherenkov light emission which is detected by the 3-D array of PMTs. The PMT position and the arrival time of the light allow the reconstruction of the lepton (muon) track. NTs are built very deep underground/undersea in order to reduce as much as possible the atmospheric muon flux. γ c CR interaction with the atmosphere µ θ c = 42º ν 4
The ANTARES neutrino telescope (see Circella s talk) Storey Buoy Electronics container Horizontal layout PMT 350 m 40 PMT km electro-optical cable to shore 12 lines 25 storey/line 3 PMT/storey 884 PMTs in total 40 km off Toulon coast (France) 2500 m depth 42º50 N 6º10 E 60-75 m 12 lines 3x25 PMT/line Submersible Detector Anchor completed on May 2008: 12 lines deployed (+1 instrumentation) 100 m Interlink cables 5 PMT Junction box
A neutrino candidate The ANTARES detector Example of a reconstructed up-going muon (i.e. a neutrino candidate) 6
Track reconstruction The ANTARES detector Reconstruction of µ trajectory from time, charge and position of PMT hits Track reconstruction method in two main steps: 1) Linear pre-fit: first estimation of the track parameters is performed 2) Final fit (ML method): PDF function of hit time residuals ( t) includes the full knowledge of the detector and the expected physics. Quality cut of the reconstruction Declination distribution of log(l) Λ + 0.1( both real data and MC for comp 1) elevation < -10 && Λ > -4.7 Log-likelihood per degree of freedom DOF Number of compatible solutions 94 events Good agreement between real data and MC Time residuals Λ best 7
2007 data (5-line) detector performance Neutrino effective area for the 2007 period, averaged over the neutrino angle direction. Selection of different nadir angles (Φ) evidences the Earth opacity at higher energies The ANTARES detector Neutrino angular resolution (angle between the true neutrino and the reconstructed track) for the 2007 period. Very good angular resolution, crucial for point-like sources search A eff ν ~ 4 10-2 m 2 @ 10 TeV The angular resolution is better than 0.5 at high energies (E ν > 10 TeV) 8
Methods of point-like sources search The background from atmospheric neutrinos and muons will be dominant. It is crucial to have an algorithm able to point out the accumulation of the signal events over this background. The background is right ascension independent and declination dependent. ANTARES: Very good angular resolution: < 0.3º for Eν > 10 TeV. Sources are visible up to δ = 47º. GC is visible (63% of the time). Background-like Signal-like Binned methods: Grid (square shape). Cone (circular shape). Unbinned methods: ML ratio. Expectation- Maximization (EM). δ = 47º 9??
BIN: Cone method In the all sky search each event is taken as the cone centre. In a fixed-source search the source position is taken as the cone centre. The cone size is optimized to get the best signal/background ratio: MDF (all sky search) and MRF (fixed-source search). The probability for the background to produce a given number of events can be computed analytically. δ RA P i is the probability for the background to produce the observed number of events N 0 or more (up to the maximum number N total ). σ is each element of the set C Ntotal n of combinations of N total elements in groups of n elements. Probability for the event j to be inside the cone defined for the event i. 10
Signal & Background BIN: Cone method The number of background events (nb) inside the cone is estimated from the real data. declination (rad) The PSF is obtained from MC assuming a flux of E -2. 11
Cone size optimization BIN: Cone method nobs (nb) µ (nb) = 90 µ 90(nobs,nb) exp( nb) (n )! n obs obs Upper limit Poisson weight MRF δ = -30 r best ~ 3 Optimum radius (deg) MRF = Model Discovery Factor used for all-sky search Model Rejection Factor used for fixed-source search MDF δ = -15 r best ~ 2.5 cone radius (deg) Optimum radius (deg) declination (deg) cone radius (deg) declination (deg) 12
UNBIN: EM algorithm The Expectation-Maximization (EM) is a pattern recognition algorithm that analytically maximizes the likelihood in finite mixture models. Mixture models: different groups of data are described by different density components. g = number of mixture models π i = mixture proportions, where g j = 1 π j = 1 Previous step: change from incomplete to complete data set. { x} x i = ( α ra i, δ ) i { y} y i = ( α ra i, δ, z i i ) The vector z i is a class indicator that indicates if the event i belongs to the background or the source. Expectation step Start with a set of initial parameters Ψ (m) = {π 1,π 2,µ,Σ} Expectation of the complete data log-likelihood, conditional on the observed data {x} m) Q( Ψ, Ψ ) = E[log( g({ y}; Ψ)) p({ x}; Ψ ( ( m) Maximization step Find Ψ = Ψ(m + 1) that maximizes Q(Ψ, Ψ(m)) 13 )]
Signal & background UNBIN: EM algorithm A simple pre-clustering selects a set of candidates by using a cone of 5º around each event of the sample. Background from real RAscrambled data Signal pdf model Background Signal Sample 2D-Gaussian p( x) = π P ( δ ) + π P ( x; µ, Σ) BG BG S S position of event: x = (α RA, δ) bg: only δ signal : α RA, δ Samples simulation: 10 4 samples simulated. Each sample corresponds to the lifetime of 2007 data period. The Bayesian Information Criterion (BIC) is chosen as the test statistic of the method. BIC k Likelihood ratio = 2logp( D Ψˆ k, M k ) v log( n) Background like 14 k penalty Signal like 10 4 samples
RESULTS 15
Fixed-source search RESULTS 24 sources in the ANTARES field of view have been selected among the most promising neutrino source candidates (galactic and extragalactic) for the 2007 data (5Line) point-like source analysis. Sky map in galactic coordinates 24 selected sources + IceCube Hot Spot (δ= 11 α = 153 ) Sky coverage of 3% 16
Fixed-source search RESULTS The p-value is the probability of the background to produce the measured (or higher) observable (BIC for the EM algorithm or n events for the con method). The P-values 1 lowestfound value for 4 sources corresponds in our tolist. a p-value pre-trial of 2.8σ found with UNBINNED method. ItPis expected in 10% of the experiments when looking at 25 sources (post-trial probability). # events inside the cone Probability distribution of the background when we look at the 25 positions in the sky: the probability The low p-value is due to to find 4 (or more) the position of the event events very close in our to the cones source is about location. 33% 17
Fixed-source search RESULTS Upper limits obtained with 2007 data (5 lines, 140 livetime days), compared with 1 year of complete detector (12 lines) and other experiments. 18
All-sky search RESULTS Sky map with the 94 events selected for point-like source analysis with 2007 data. EQUATORIAL COORDINATES First neutrino sky map of ANTARES 19
All-sky search RESULTS EM algorithm BIC distribution of only background BIC obs In our sample : BIC obs = 1.4 (highest value) p-value = 0.3 (1σ excess) (δ = -63.7º RA =243.9º) No significant excess was found Cross-check: Cone method δ RA N events P BIN -28.8 31 1 0.053-42.7 164 1 0.050 24.2 32 1 0.069-51.4 159 1 0.052-63.38 244 1 0.055 20
Summary The ANTARES detector is completed and observing neutrinos steadily. During the 5-line period (Feb 2007 Dec 2007) 94 neutrino events were selected for a point-like sources search. Two methods: one binned (cone method), and another unbinned (EM algorithm) have been used for the analysis. Neither the fixed-source search nor the all sky survey have revealed any significant signal excess in the ANTARES field of view. The analysis of the data from 2008 with the full detector operating is in progress. 21