Abstract
Cataclysmic events such as the merger of black holes and neutron stars can lead to the emission of gravitational waves. Since 2015, around a hundred such signals have been detected. As the detectors are upgraded, more signals will be detected, opening the door to new scientific avenues. One such avenue
... read more
is the strong lensing of gravitational waves, where the presence of a massive object along the wave’s trajectory leads to it being split into several images with the same frequency evolution. In this thesis, we demonstrate a new and faster method to search for and analyze such lensed signals. In addition, we show two different avenues to decrease the high false-alarm probability related to lensing searches. One uses higher-order modes to identify signatures intrinsic to lensed gravitational-wave signals. The other consists of including the expected distribution for the lensing parameters in the detection statistic, effectively reducing the false alarm. Besides lensing, upgraded detectors also offer the possibility to accumulate information about the signal earlier before the merger. This is important when one wants to jointly analyze gravitational wave and electromagnetic data for binary neutron star signals. In this thesis, we exploit machine-learning-based techniques to set up an early alert system able to issue alerts up to a minute and a half before the merger happens. In parallel, we develop a system to rapidly generate sky maps for detected signals. Even if they are currently at the stage of proof-of-concept, joining these two approaches could make for the first full machine-learning-based early-alert system. In addition to upgrading the existing detectors, the next generation of ground-based detectors is already planned. These detectors should see many more events and have an increased sensitive band, making for long signals. Therefore, signals could start overlapping. In this thesis, we establish that overlaps will be common for these detectors. Then, we study their impact on data analysis and establish that, in some cases, not accounting for the overlapping signals can lead to erroneous results. This motivates the development of new analysis techniques able to deal with this observation scenario. In this thesis, we suggest two different methods: hierarchical subtraction and joint parameter estimation. The first is faster than the second but more keen on biases due to the overlapping signals. An issue with these methods is their speed as they would be unable to keep up the pace with the expected detection rate. Therefore, in this thesis, we also explore the possibility to analyze overlapping signals using machine learning techniques. While slightly less precise, our method is able to analyze overlapped binary black hole signals in about a second, compared to 20 to 30 days for Bayesian methods. Further upgrades to our framework should make it able to analyze joint signals, opening the door to more precise simulation studies for next-generation detectors.
show less
