This project will further the development of a new computer code for cosmological N-body simulations. Abacus implements a novel method for computing gravitational forces that is a hundred times faster and substantially more accurate than current methods. It can be run on commodity graphics processing units and offers a vast improvement in price-performance. Observational surveys of large scale structure rely on cosmological simulations to calibrate their statistical estimators and compute accurate error bars, making it essential to run large grids of highly accurate simulations in order to realize the potential of current and future cosmological surveys. Abacus relies on a mathematical breakthrough that accomplishes a clean split between the near-field and far-field force on particles. The present work will add essential support for individual particle time steps, on-the-fly group finding and merger tree building, and light-cone outputs. The study will also extend Abacus to a parallel implementation, and allow release of a public version of the code.
The research team will use Abacus for a first suite of large simulations, whose resulting halo catalogs, merger trees, and light-cone outputs will be made public as a direct benefit to current and future survey projects. Abacus is a game-changer, opening up the ability to do large simulations to many more astrophysicists at exactly the time that the need for large sets of large simulations will intensify. The mathematical approach of Abacus opens up substantial new possibilities in the solution of elliptic equations, one of the most common computational tasks across a wide range of scientific and engineering disciplines. Along the way, the project will train a post-doctoral researcher and a graduate student.
