Content area
Abstract
Numerical relativity, the direct numerical integration of the Einstein field equations, is now a mature subfield of computational physics, playing a critical role in the generation of signal templates for comparison with data from ground-based gravitational wave detectors. The application of numerical relativity techniques to new problems is at present complicated by long wallclock times and intricate code. In this thesis we lay groundwork to improve this situation by presenting a GPU port of the numerical relativity code SpEC. Our port keeps code maintenance feasible by relying on various layers of automation, and achieves high performance across a variety of GPUs. We secondly introduce a C++ software package, TLoops, which allows numerical manipulation of tensors using single-line C++ source-code expressions resembling familiar tensor calculus notation. These expressions may be compiled and executed immediately, but also can be used to automatically generate equivalent GPU or low-level CPU code, which then executes in their place. The GPU code in particular achieves near-peak performance. Finally, we present simulations of eccentric binary black holes. We develop new methods to extract the fundamental frequencies of these systems. Using these frequencies we identify when these binaries pass through coordinate resonances, at which points high mass-ratio inspirals can experience short-timescale phase-dependent deviations from smooth inspiral called "kicks". We find no evidence for such kicks at comparable mass-ratio.
