Gravitational-wave astronomy: modelling, detection, and data analysis

• Autores: Alejandro Torres Forné
• Directores de la Tesis: José Antonio Font Roda (dir. tes.) , Pablo Cerdá Durán (codir. tes.)
• Lectura: En la Universitat de València ( España ) en 2017
• Idioma: español
• Tribunal Calificador de la Tesis: Antonio Marquina Vila (presid.) , Nikolaos Stergioulas (secret.) , Giovanni Prodi (voc.)
• Enlaces
  • Tesis en acceso abierto en: TESEO
• Resumen

  • This thesis is divided in three main topics of research related with gravitational-wave astronomy, namely (i) on the numerical modelling of astrophysical sources of gravitational waves, (ii) on the detector characterization and classification of transients of noise, and (iii) on the development of new methods for data analysis.

    In the context of numerical relativity, this thesis includes the simulation of the accretion processes onto neutron stars to study the so-called hidden magnetic field scenario. I have developed a 1D model and performed a parameter-space study aimed at determining under which conditions this scenario can be a viable model to explain the low magnetic fields observed in some central compact objects in supernova remnants. The cause of those low values of the magnetic field is still unclear. Previous numerical simulations suggested that the magnetic field can be compressed to the surface of the star due to the pressure exerted by the infalling fluid. Our results show that the accretion rate required to compress the magnetosphere of a typical pulsar is modest. Therefore, this scenario should not be regarded as particularly unusual. However, our results also show that it is fairly complicated to compress the magnetic field if it is stronger than $10^{14}$ G, which are the typical values found in magnetars, since the required accretion rate would cause the star to collapse to a black hole.

    To this day, the physical mechanisms behind core-collapse supernovae explosions and their subsequent evolution are still not entirely known. The information contained in the gravitational-wave signal produced by this type of sources can greatly help determining the physics involved in the explosions. This thesis presents the first results of a project aimed at inferring some of those physical parameters from the study of gravitational-wave signals from core-collapse. I have studied the existing relationship between the modes of oscillation of the proto-neutron star that forms after the collapse of a massive star and the spectrum of the gravitational signal. The model handles the oscillations of the proto-neutron star as the perturbations of a system in equilibrium. We compare the data from the gravitational-wave signal generated by the simulation of the collapse of a massive star with the time-frequency distribution of the different modes of oscillation, obtaining a remarkably close correspondence between them.

    Detectors of gravitational waves are affected by many sources of noise due to the extreme sensitivity required to measure the small-amplitude variations caused in the distance between test masses by passing gravitational waves. The search for the sources of noise and their subsequent elimination is a fundamental task. This thesis presents results of a collaborative project to automatically classify and remove noise transients (glitches) produced in the advanced LIGO and Virgo detectors. Some of these transients may be particularly problematic because they can be misinterpreted as true gravitational-wave signals. The three methods employed in the project are able to correctly classify 95\% of detected glitches. Since all three methods use different strategies to perform the classification, they are complementary, so that transients not classified by one of the methods may be classified by the other two.

    In spite of the efforts to reduce the noise of interferometers, it is inevitable that part of the noise may affect and bury actual gravitational-wave signals. There are many data-analysis methods designed to extract signals from noisy backgrounds. During this thesis, I have explored the performance of denoising algorithms based on the concept of Total Variation. These algorithms, which do not require any a priori information about the signal, have been shown to be highly efficient for noise suppression in the context of image processing. Our pioneering results for gravitational-wave signals show that the algorithms can remove enough noise to produce distinguishable signals in the two scenarios we have considered, signals mixed with Gaussian noise and with real detector noise. One of the most interesting future applications of this line of work is the combination of these methods with other common techniques of gravitational-wave analysis (e.g.~Bayesian inference) to improve the results. Finally, in an attempt to bridge the gap between numerical modelling and data analysis, this thesis has also explored the use of dictionary-learning techniques with numerical-relativity waveform templates in order to reconstruct signals embedded in Gaussian noise. These techniques offer a number of possibilities, not only to extract signals from noise, but also to classify glitches or to extract physical parameters from detected signals.