Brief introduction
Experimental data and simulated samples
Trigger level
Levels 1 & 2 (direct walk & likelihood
reconstruction)
Level 3 (Quality cuts)
Neural network selection
Final cuts and events selection
Expected limits
Summary
Comparison of 2000 and 2001 analysis
We present recent results from a search for neutralino dark matter with AMANDA-II based on the data taken during year 2000.
Further information concerns the results obtained on the blind 100% of 2000 AMANDA data.
Experimental data set
Corresponds to 196.0 days (1.41·109 triggers) of the detector livetime (runs 197-584). The detector deadtime in year 2000 is about 17.2% of the detector operational time. In accordance with file-wise re-triggering/cleaning procedure three different time periods were defined:
The 2000 data subset when the Sun is below the horizon corresponds to 137.0 days (about 0.38 year) of the detector livetime (runs 231-495) with estimated detector deadtime of 17.0%.
Atmospheric muon simulation
Atmospheric showers were generated using dCORSIKA (v6.020) with QGSJET interaction model. Simulated sample of atmospheric muon background corresponds to 28.5. days.
Atmospheric neutrino simulation
Atmospheric neutrinos were simulated with a neutrino generator nusim. The energy and angular distribution of the primaries were obtained by re-weighting of the generated events sampled from an E-1 spectrum.
Atmospheric neutrinos (100 files with 10000 generated events for each year) were simulated with energies between 10 and 108 GeV and zenith angles between 80° and 180° (in the AMANDA coordinate system).
Neutralino signal simulation
In this work the program WimpSimp (v1.2) was used as a neutralino-induced neutrino generator. Two extreme neutralino annihilation channels were chosen:
The signal simulation has been performed for both annihilation channels and neutralino masses 50, 100, 250, 500, 1000, 3000 and 5000 GeV. These values roughly correspond to the different theoretical predictions. Each simulated sample of neutralino-induced neutrinos consists of 1500000 events.
To simulate neutrino-nucleon interactions in the detector vicinity the code GENNIUP based on PYTHIA was used.
Muon propagation & detector response
Simulated muons were propagated through the ice to the detector active volume with MMC (v1.09).
The propagation of photons inside the full active detector volume was performed using pre-calculated photon tables for MAM ice model.
The final step in the simulation (the detector response to the considered muon event) is simulated using mandarin-1.1 release of AMASIM (v2.73.14).
An overview of the performed event selection can be found in Table 1
Pre-cleaning & filtering
Before the actual reconstruction was performed, the data passed several cleaning steps and timing and amplitude calibrations. First step in the pre-processing includes selection of events flagged by the AMANDA-II majority trigger. This corresponds to about 90% of all triggers. After the trigger event selection, the hit cleaning and re-triggering are performed.
Effective volumes & areas
Effective volumes & areas for neutralino-induced muons at the trigger level (Veff and Aeff) are plotted as a function of the expected neutralino masses.
Level 1 is based on direct_walk-II first guess. Only events with reconstructed zenith angles larger than 70° are kept. Then, reconstructed zenith angles are used as input for Pandel likelihood reconstruction.
At the level 2 the final cut on the reconstructed likelihood zenith angle at 80° is added.
Passing rates after Levels 1 & 2 can be found in Table 3.
At the level 3 following cuts were applied:
Cut values of these variables were chosen by optimizing model rejection potential or/and the best possible signal-to-background ratio.
Since the generation region of neutralino-induced muons is strongly defined by the Sun location (generated angles are confined between 90° and 113.5°), it seems reasonable to apply cuts on reconstructed zenith angles around this region.
Selected cut values are listed in Table 2.
Passing rates after Level 3 cuts can be found in Table 3.
A final step in the signal separation was based on neural network predictions. The neural net package MLPfit provided with PAW was used.
The constructed network represents a multi-layer perceptron (feed-forward network) and consists of 7 input neurons, 1 hidden layer with 35 neurons and 1 output neuron. To avoid a network's overfeeding, the training was limited to 250 epochs. The network was trained with a desired output value equal 1 for signal events (from neutralino annihilations) and 0 for background events (simulated atmospheric muons). Sizes of training subsamples were taken equal to approximately a fifth part of corresponding muon sample. No information about simulated atmospheric neutrinos was given to the network during training.
The networks input variables are listed below:
For the final event selection the network output, Q was used as a cut parameter. Cut values depend mainly on neutralino masses.
Few additional cuts on other variables were applied to remove remaining atmospheric neutrino events. Extra cuts are listed below:
Quantitative values of applied cuts are listed in Table 4.
After the final cuts all simulated atmospheric muon events were removed. An extrapolation of background muon sample to the operational detector time taking into account passing rates at the final analysis levels as well as a visual inspection of surviving events indicates that there are no atmospheric muon background events expected among selected events.
The shape of final distributions -- number of hit channels, Nch; reconstructed zenith angles, Zenith(pl); and length of direct hits type B, LdirB -- have been tested using Kolmogorov-Smirnov test (.ps). The agreement in shape is within statistics.
Distributions (Zenith vs. Azimuth) for experimental data (100 GeV filter, 500 GeV filter) show a certain structure which reflects a geometrical features of the detector. At the same time these distributions have no obvious clustering.
The angular resolution -- zeniths (500 GeV filter) and azimuths (500 GeV filter) -- of neutralino-induced events passed the final selection depends on muon energy and is about 3.5° for neutralino with m>250 GeV.
Final effective volumes (.eps) and areas (.eps) for neutralino-induced muons are plotted as a function of the expected neutralino masses.
Passing rates after neural network selection can be found in Table 3.
The number of data and atmospheric neutrino background events after the final cuts presented in Table 5.
After the final event selection no excess of the data over expected atmospheric neutrinos in the selected zenith angle range was observed. Due to blindness requirements the Sun position is still unknown.
An upper limit on the detector sensitivity to the neutralino-like signal in the selected zenith regions can be derived from the upper limit on the expected number of signal events passed the final level. From the upper limits on the expected neutralino-like signal, mus and corresponding effective volumes for the simulated muon samples one can set 90% CL upper limits on the neutrino to muon conversion rate close to the detector. The conversion rate is strongly model dependent (since the unique effective volume is defined for each neutralino mass and annihilation channel).
The calculated upper limits on the neutrinos to muons conversion rates near the detector can be converted into a corresponding upper limit on the neutralino annihilation rate in the Sun.
The calculated limit on the expected detector sensitivity to the neutralino-like muons shows that AMANDA-II limit for 2000 data (.eps) (0.38 years of the detector livetime) are compatible with SuperK and Baksan experiments.
Table 1 Summary of the different processing levels;
Table 2 L3 cut values;
Table 3 Passing rates at the different levels;
Table 4 Cuts on NN output parameter and extra cuts (L5);
Table 5 Number of data and atmospheric neutrino background events after L5;
In this analysis we tried to develope a more or less unified approach for both 2000 and 2001 years. (NB. The 2001 data was unblined last spring). Here are a few comments on 2000 and 2001 filtering:
| Back to main page | Yu. Minaeva
30/IX-2004 |
