Simona Toscano

1
Point-like source searches with the ANTARES
neutrino telescope
TeV Particle Astrophysics IV 24-28 September
2008, IHEP, Beijing

• Simona Toscano
• IFIC (Instituto de Física Corpuscular)
  CSIC-Universitat de València, Spain
• on behalf of the ANTARES collaboration

2
Neutrino telescopes
Scientific scope of a Cherenkov Neutrino
Telescope
?
MeV GeV-100 GeV GeV-TeV TeV-PeV PeV-EeV gt EeV
ANTARES is a powerful tool to search for
neutrino point like sources
12-Line detector
Angular resolution better than 0.3 above a few
TeV
Galactic Centre visible 63 of time
Sky coverage in Galactic coordinates for a
detector located in the Mediterranean Sea and at
the South Pole. The locations of recently
observed sources of very high energy (VHE)
??-rays are also indicated.
3
ANTARES detector
(See Zornozas talk in plenary session for
details.)
Positions of reconstructed tracks at time of
first triggered hit
a neutrino-induced muon crossing the detector
Footprint of the 12-line detector in atmospheric
muons
4
Methods for the search of point-like sources
Different methods have been developed within
ANTARES collaboration for the search of
point-like sources
5
CONE METHOD
EM ALGORITHM
Source position

• Sky is divided in a grid of bins (to perform a
  full-sky survey) or cones around the source
  position (for a fixed-source search).
• The optimum size of the bins/cones is calculated
  for maximum sensitivity.

The optimum search cone radius (calculated for
each d) corresponds to the minimum MRF (model
rejection factor) .
MRF is defined as
The average upper limit (aka sensitivity) is
calculated assuming that no true signal is
present (ns0) and only the expected background
(nb) of natm and matm is observed. For a Poisson
distribution of the background, the average upper
limit is
Poisson weight
Upper limit
6
CONE METHOD
EM ALGORITHM
The EM method is a pattern recognition algorithm
that analytically maximizes the likelihood in
finite mixture problems, which are described by
different density components (pdf).
mixing proportions
For a point-like source
J.A. Aguilar J.J Hernández. Astroparticle
Physics doi10.1016/j.astropartphys.2007.12.002
7
Analysis of ANTARES 5-Line data
5-Line data from Jan to Dec 2007
Real data Silver (Ag) Runs equivalent to 140
days live time (Ag baseline lt 120 kHz burstfr
lt 40) MC data

• Neutrinos
• 130 files of n and anti-n
• Spectrum of generated n E-1.4
• Energy range 10 107 GeV

x,y track positions at time of first triggered
hit

• Muons simulated with CORSIKA
• Primary ions -gt p, He, N, Mg, Fe
• Primary energy -gt 1 ? 105 TeV/nucleon
• Primary zenith angles gt 0? 85
• Primary spectrum ? E-2
• Number of simulated showers 1010
• Live time -gt hours (or days) years (depending
  on the mass, energy and angle)

First neutrino with 5-Lines
8
Reconstruction algorithm
Reconstruction of m trajectory from time, charge
and position of PMT hits

• Track reconstruction method in two main steps
• Linear pre-fit first estimation of the track
  parameters is performed
• Final fit (ML method) PDF function of hit time
  residuals (Dt) includes the full knowledge of
  the detector and the expected physics.

Quality cut of the reconstruction
Declination distribution of both real data and MC
for elevation lt -10 L gt -4.7
Log-likelihood per degree of freedom
Number of compatible solutions
9
5-Line detector performance
Selection of different nadir angles (F) evidences
the Earth opacity at higher energies.
The angular resolution is better than 0.5 at
high energies (En gt 10 TeV)
Aneff 410-2 _at_ 10 TeV
10
Background estimation

• A fit of d distribution, for given quality cuts,
  is performed from MC or real data

1. PBG(d) fit from MC or real data gives the
   background pdf used in the algorithm.
2. Samples simulation

• Sample simulation
• 104 samples simulated.
• Each sample corresponds to 140 days (5-Line
  detector live time).

Signal
The background inside the cone, for increasing
cone size, is estimated for a given declination.
11
Signal simulation
Neutrino angular resolution
Neutrino (MC) angular error distribution for
different declination bands for a spectral index
of 2
median angle between the true neutrino track
(from MC) and the reconstructed track.
CONE METHOD
EM ALGORITHM

• The signal inside the cone is calculated from the
  angular error distribution

The signal simulation has been done using
the angular error distribution
A declination band and the desirable number of
events are selected.
Angular distances around the source location are
randomized according to the angular error
distribution
12
Optimization of the search cone radius
CONE METHOD
EM ALGORITHM
The cone which minimizes the MRF is the optimum
cone for point-like sources search
Optimal cone radius for any declination.
MRF as a function of cone radius for a given
declination
Expected background and fraction of signals in
the cone as a function of declination
d -30 rmin 3
13
Antares 5-Line Sensitivity
ANTARES 5-Line sensitivity compared with the
results presented by other neutrino experiments.

• Sensitivity in the integrated neutrino flux
  (above En 10 GeV) for a spectral index of 2.
• The average increase of the unbinned method over
  the binned method is about 27.

14
Conclusions
LAST NEWS from ANTARES meeting Unblinding
proposal approved for the 5-Line analysis
15
ANTARES(12-Line) sensitivity
ANTARES expected sensitivity in one year of
data-taking . We can compare the result in terms
of sensitivity with respect to different
experimental results and projected performances
of several neutrino experiments.
16
Sensitivities

• Unbinned 5-Line vs 12-Line with different quality
  cuts

Factor 10
Factor 7
17
CONE METHOD
EM ALGORITHM
The EM method is a pattern recognition algorithm
that analytically maximizes the likelihood in
finite mixture problems, which are described by
different density components (pdf) as
Point-like sources

• We assumed only one source, g1
• Background only depends on declination
• The reconstructed energy is not used

1. The background pdf is extracted from MC data or
   real RA-scrambled data when available
2. Signal pdf model is selected to be 2D-Gaussians

18
General procedure
CONE METHOD
EM ALGORITHM
The idea is to assume that the set of
observations forms a set of incomplete data
vectors. The unknown information is whether the
observed event belongs to a component or another.
Easily differentiable!

• E-Step (Expectation-step)
• Start with a set of initial parameters ?(m)
  p1,p2,µ,S
• Expectation of the complete data log-likelihood,
  conditional on the observed data x
• M-Step (Maximization-step)
• Find ? ?(m 1) that maximizes Q(?, ?(m))

Successive maximizations of the function
Q(?,?(m)) lead to the maximization of the
log-likelihood
19
Searching procedure
CONE METHOD
EM ALGORITHM
20
Model Selection
CONE METHOD
EM ALGORITHM
We use the model testing theory to calculate the
significances. As a test statistic we use the
Bayesian Information Criterion (BIC)
Discovery power
In the case of two model testing
(Only-background, M0, and backgroundsignal, M1)
is given by
Confidence level
Likelihood ratio
penalty
Although it sounds bayesian is used in a
frequentist fashion.