Lab researchers help develop power spectrum emulator code for cosmology studies

August 18, 2010—Many of the most exciting questions in cosmology, including observational probes of dark energy, rely on an understanding of the nonlinear regime of cosmic structure formation. In order to exploit the information available from this regime and to extract cosmological constraints, accurate theoretical predictions are needed. Currently, expensive numerical simulations are required to produce accurate predictions for structure formation from differing cosmological parameters. Fast prediction tools, called emulators, based on a relatively small number of high-precision simulations, could replace the simulator in the analysis work.

LANL scientists and collaborators have produced a prediction scheme that gives accurate results in a fraction of the time for a range of parameter values. For a model for the Universe built largely of cold dark matter with variations in dark energy, the researchers generated predictions from a set of cosmologies spanning the parameter space of interest. The scientists employed the Coyote Universe simulation suite that comprises 1,000 simulations, requiring on the order of a million CPU-hours, to obtain high-performance precision simulations of the power spectra. Each simulation models the Universe with approximately a billion particles whose positions evolve according to gravitational calculations. In each simulation, the scientists computed the matter power spectrum to describe the density of the Universe at different scales. The researchers combined replicate spectra using a statistical technique to estimate the unknown smooth spectrum. The smooth estimates are used to build the prediction framework. This framework, called an emulator, is a function that takes a set of cosmological parameters and produces a prediction for the power spectrum. The emulator is a statistical model trained on the estimates from the previous step.

The work demonstrates that it is possible to extract cosmological statistics, such as the power spectrum, at high accuracy and to build an accurate prediction scheme based on a limited set of simulations. The scientists publicly released the CosmicEmu emulator to interpret large cosmological surveys. The Astrophysical Journal published the work.

Researchers include Earl Lawrence, Dave Higdon, and Brian Williams of Statistical Sciences (CCS-6), Katrin Heitmann of Space Science and Applications (ISR-1), Salman Habib of Nuclear and Particle Physics, Astrophysics and Cosmology (T-2), Martin White of the University of California, Berkeley, and Christian Wagner of Astrophysikalisches Institut Potsdam and Universitat de Barcelona. The LANL Institutional Computing Initiative and the Laboratory Directed Research and Development (LDRD) supported the work at Los Alamos.