Skip to content

Latest commit

 

History

History
120 lines (64 loc) · 6.86 KB

README.md

File metadata and controls

120 lines (64 loc) · 6.86 KB
Raw

STACKED SEARCH FOR CORRELATION BETWEEN ICECUBE NEUTRINOS AND RADIO PULSARS

This repository contains the code used to perform the analysis described in the paper "A stacked search for spatial coincidences between IceCube neutrinos and radio pulsars" (https://arxiv.org/abs/2306.03427). The code is written in Python 3.10 and uses the following packages: numpy, scipy, matplotlib, pandas, numba, multiprocessing. The code is provided as is, without any warranty. If you use this code, please cite the paper.

Description of the code

The repository is organized as follows:

- core
- data
- outputs

core

The core/ directory contains the following files:

  • download_ATNF.py : downloads the ATNF pulsar catalogue and saves it in data/ATNF.txt
  • download_IC.py : downloads the IceCube neutrino data package and saves it in data/icecube_10year_ps
  • stacking_analysis.py : defines the functions used to calculate the Test statistic, and the signal flux model and other miscellaneous functions used in the analysis indicated in the paper https://arxiv.org/abs/2306.03427
  • req_arrays.py : Stores the required data in the form of numpy arrays for faster and easier computation
  • readfiles.py : Reads and refines the data from the files and stores them in the form of panda.Dataframes
  • signal_bag.py : (depercated) defines the functions used to compute the Signal and Background PDF in the analysis indicated in the paper https://arxiv.org/abs/2306.03427
  • weights.py : (depercated) defines the functions used to compute the weights of pulsars for the analysis indicated in the paper https://arxiv.org/abs/2306.03427

data

The data/ directory contains the data used in the analysis. The data are taken from the following sources:

outputs

The outputs/ directory contains the results of the analysis in the form of images.

Computational Complexity

The signal PDF requires heavy computation which consists:

  • Angles between 2374 pulsars * 1134450 neutrinos = 269,500,850 values.

  • $\omega_{acc}$ for 2374 pulsars. Each pulsar is attributed a weight $\omega_{acc}$ which is proportional to the declination and detector effective area. It is given by:

    $\omega_{acc,j} = T \times \int A_{eff}(E_{\nu}, \delta_j)E_{\nu}^{\Gamma} dE_{\nu}$

    where $T$ is the livetime of the detector, $A_{eff}$ is the effective area of the detector, $\Gamma$ is the spectral index of the neutrino flux, and $\delta_j$ is the declination of the $j^{th}$ pulsar. The effective area of the detector is a function of the neutrino energy and the declination of the source. The effective area is calculated for 1134450 neutrinos and 2374 pulsars.

  • The above integral is calculated using np.trapz for 1e7 energies for each of the 10 seasons of the IceCube and for 3 different spectral indices for 2374 pulsars.

  • The no.of signal events for each pulsar indexed by $j$ is given by:

$\hat{n}{s_j} = T \times \intop A{eff}(E_{\nu}, \delta_j)\dfrac{dF}{dE_{\nu}}dE_{\nu}$

$T$, $A_{eff}$, and $\delta_j$ are the same as above. $\dfrac{dF}{dE_{\nu}}$ is the expected neutrino spectrum from the source and can be modelled using a power-law as follows:

$\dfrac{dF}{dE_{\nu}} = \phi_0  \left( \dfrac{E_{\nu}}{100 \text{ TeV}}\right)^{\Gamma}$

where  $\phi_0$ is the flux normalization and $\Gamma$ is the spectral index. Each $\hat{n}_s$ is calculated for 1e7 energies for each of the 10 seasons of the IceCube and for 3 different spectral indices for 2374 pulsars. 
<!-- This requires computing 2374 * 1e7 * 30 = 7.122e11 values. -->
  • The above two integrals are calculated using np.trapz for 1e7 energies for each of the 10 seasons of the IceCube and for 3 different spectral indices for 2374 pulsars. This requires computing 2374 * 1e7 * 30 = 7.122e11 values.

Then the stacked Signal PDF for each neutrino $i$ is calculated by taking the weighted sum over the individual pulsars with three different weight models.

The Likelihood of $\hat{n}_s$ signal events is given by:

$\mathcal{L}(n_s) = \prod_i \frac{n_s}{N} S_i + (1-\frac{n_s}{N}) B_i$

where $N$ is the total number of neutrinos, $S_i$ is the signal PDF for the $i^{th}$ neutrino, and $B_i$ is the background PDF for the $i^{th}$ neutrino.

The background PDF is the no.of neutrinos within 5$^\circ$ declination band of the $i^{th}$ neutrino. The background PDF is calculated for 1134450 neutrinos.

The Likelihood is calculated for 3 different spectral indices.

We then easily calculate upper limits by using scipy.interp1d as the no.of values required are less

Solution to the computational challenges

  • This is a very large number of values to compute. To overcome this challenge, we use the following techniques:

  • Use the numba.njit package to speed up the computation. The numba package is used to compile the python code into machine code.

  • The integrals are evaluated using functions accelerated by numba.vectorize package and np.trapz, which is faster than scipy.quad or scipy.dblquad.

  • Used numba.vectorize package to replace a for loop calling functions like ns_sing_season, psr_wt_acc. This reduce the computation time by a factor of 10.

  • Used multiprocessing to parallelize the computation. The multiprocessing package is used to run the code in parallel on multiple cores of the CPU.

  • Running the code normally requires ~ 2e12 calculations which take > 2 days of continuous computation.

  • Using the above techniques, the computation time is reduced to ~ 2 hours.