EFT and AI Techniques for Gravitational Waves Physics

This thesis presents synergistic advancements in Effective Field Theory (EFT) and Artificial Intelligence (AI) techniques applied to gravitational wave (GW) physics. An extension of the EFT of Gravity for Extended Objects was developed to incorporate parity-violating effects, driven by the quest for exotic compact objects and fundamental symmetries. This formalism enabled the calculation of GW signals from binary systems encompassing finite size, spin, and dissipation, particularly highlighting unique signatures arising from parity violation. These include distinct Post-Newtonian ordering of radiating moments, novel spin-induced orbital dynamics, characteristic dissipative forces, and specific GW polarization content. The study also advanced the parity-conserving sector by computing the gravitomagnetic current quadrupole moment. Additionally, to model environmental influences on binary inspirals, a novel Effective Field Theory was developed, capable of systematically incorporating the finite-size effects of objects moving within fluids into a computationally tractable point particle formalism. This approach circumvents the need for complex dynamical boundary conditions. The fundamental nature of this framework was demonstrated through a first-principles derivation of d’Alembert’s paradox. Key applications included the calculation of the generalized Einstein-Infeld-Hoffmann potential for binaries immersed in a perfect fluid. The formalism was further extended to encompass dissipative effects by devel?oping a novel, generally covariant treatment of viscosity using the Schwinger-Keldysh framework, resulting in a relativistic generalization of Stokes’ law. The pursuit of detecting subtle GW effects revealed limitations in conventional analysis paradigms, particularly the dependence on predefined waveform templates for matched filtering and its sensitivity to complex detector noise. To overcome these challenges, novel AI methodologies were developed. An approach utilizing transfer learning demonstrated the ability for AI models to rapidly adapt to non-Gaussian and time-varying noise, enhancing the robustness and real-time capability of GW searches. Furthermore, recognizing the inherent bias of template banks, "seeing only what we expect", a generative AI model was developed to create novel gravitational waveforms, potentially capturing missing physics and significantly accelerating waveform generation compared to traditional simulations.