Emulating the Matter Power Spectrum for the Universe
Answering many of the most exciting questions in astrophysics and cosmology depends on understanding the formation of large-scale structure in the Universe. In particular, accurate theoretical predictions of large-scale structure properties are required in order to constrain cosmological models to agree with observed data. Currently such predictions can only be obtained from costly numerical simulations. In the first part of the talk, I will discuss how to construct an accurate predictor for the matter power spectrum (an important summary of structure) using simulations for just 37 parameter settings. Each simulation produces a noisy realization of the power spectrum for its cosmology. Replications for each cosmology are combined in order to estimate the unknown smooth power spectrum for that cosmology. We treat this unknown smooth function as arising from a process convolution, a class of models built by kernel smoothing a latent random process. The estimated smooth spectra are used to construct a Gaussian process-based scheme for predicting the spectra of cosmologies that were not included in the original simulation suite. In the second part of the talk, I will discuss how we extended our predictor to include a new input parameter and wider ranges for the spectrum without the need to redo the costly simulations.