The first task we carried out was to improve and expand the (up to that point) most sensitive pipeline, BinarySkyHough. We compared the computational efficiency and sensitivity of different semicoherent detection statistics, as shown in two of the figures attached to this summary, which show that our new improved pipeline (BinarySkyHouF) is more sensitive and at the same time more computationally efficient. The main improvements were: 1) the usage of demodulated detection statistics, which allow the usage of longer coherent segments, increasing the sensitivity and giving more flexibility to the pipeline; 2) the per-detector data are combined coherently, thereby reducing the computational cost (by a factor equal to the number of detectors) and further improving sensitivity for searches with more than two detectors; 3) the new pipeline has explicit control over the mismatch in the coherent stage, allowing one to perform lower mismatch searches than before, for example, when following up an interesting candidate or targeting a particularly interesting smaller region of parameter space; 4) one can now explicitly search over binary orbital parameters such as the eccentricity and argument of periapse or higher-order frequency derivatives. These results are published in Physical Review D, available here:
https://journals.aps.org/prd/pdf/10.1103/PhysRevD.106.084035(öffnet in neuem Fenster)The second task we carried out was the development of a semi-coherent follow-up and parameter estimation pipeline using stochastic samplers such as dynesty, using the well-known and widely used bilby Python package. The main results from this project were the expansion of the capabilities of the previously available pyfstat package in a number of ways: more flexibility in the choice of sampler and prior distribution, and a new convergence criterion. We showed that for a large number of dimensions one can perform searches of CWs with a greatly reduced computational cost as compared to a search that would use a template bank (as shown in one of the attached figures). Furthermore, we showed that it is possible to find the maximum posterior point for parameter-space regions much larger than previously thought. We focused on finding a good configuration of the ptemcee, dynesty, and nessai samplers in order to reduce the computational cost of a single followup stage, showing that these samplers can produce similar results with comparable computational efficiency. We also characterized the computational cost of a parameter estimation run, and showed the improved computational efficiency of the newly developed framework. These results have been accepted for publication in Physical Review D, and for the moment are available here:
https://arxiv.org/pdf/2404.18608(öffnet in neuem Fenster)In order to achieve the second objective of our project (detect the first continuous gravitational-wave signal), we applied for computational time for a supercomputer through the PRACE platform. We obtained around 10 million hours of computing time in the Jülich supercomputer, which we have been using together with the previously discussed improved search pipeline to carry out two types of all-sky searches from unknown neutron stars in binary systems (shown in two of the attached figures): a broad search including higher frequencies with moderated sensitivity, and a narrow search with the best sensitivity ever obtained for this type of search. Unfortunately, the final results for these two searches have not yet been obtained, and will be published within the next months.
Additionally, during this project a new pipeline to carry out targeted searches has been developed, related to the previously discussed follow-up and parameter estimation development.efforts.
All of these results have been presented in multiple conferences, such as the first and second continuous gravitational waves workshops in Amsterdam and Hannover, the Spanish Astronomical Society 2022 meeting in Tenerife, the neutron stars workshop in Bonn, and the PHAROS 2022 conference in Rome.
