The upcoming generation of galaxy surveys will probe the distribution of matter in the universe with unprecedented accuracy. Measurements of the matter power spectrum at different scales and redshifts will provide stringent constraints on the cosmological parameters. However, on non-linear scales this will require an accurate evaluation of the covariance matrix. Here, we compute the covariance matrix of the 3D matter density power spectrum for the concordance $\Lambda$CDM cosmology from an ensemble of N-body simulations of the Dark Energy Universe Simulation - Parallel Universe Runs (DEUS-PUR). This consists of 12288 realisations of a $(656\,h^{-1}\,\textrm{Mpc})^3$ simulation box with $256^3$ particles. We combine this set with an auxiliary sample of 96 simulations of the same volume with $1024^3$ particles. We find N-body mass resolution effect to be an important source of systematic errors on the covariance at high redshift and small intermediate scales. We correct for this effect by introducing an empirical statistical method which provide an accurate determination of the covariance matrix over a wide range of scales including the Baryon Oscillations interval. Contrary to previous studies that used smaller N-body ensembles, we find the power spectrum distribution to significantly deviate from expectations of a Gaussian random density field at $k\gtrsim 0.25\,h\,\textrm{Mpc}^{-1}$ and $z<0.5$. This suggests that in the case of finite volume surveys an unbiased estimate of the ensemble averaged band power at these scales and redshifts may require a careful assessment of non-Gaussian errors more than previously considered.

We investigate the non-linear evolution of Baryon Acoustic Oscillations (BAO) in the low-redshift matter power spectrum from the DEUS-FUR $\Lambda$CDM model simulation. This is the first cosmological N-body simulation encompassing the full observable cosmic volume, thus allowing cosmic variance limited predictions at BAO scales. We control the effect of numerical systematic errors using a series of large volume high-resolution simulations. The combined analysis allows us to measure the matter power spectrum between $z=0$ and $1$ to 1% over the entire BAO range, $0.03<k [\textrm{h Mpc}^{-1}]<0.3$, in bins of size $\Delta k/k\lesssim$ 1%. We define the BAO with respect to a non-linearly evolved wiggle-free spectrum and determine the characteristics of the BAO without recurring to extrapolation from global fitting functions. We quantify the effects of non-linearities on the position and amplitude of the BAO extrema, and the coupling to the broadband slope of the power spectrum. We use these estimates to test non-linear predictions from semi-analytical models. Quite remarkably from the analysis of the redshift evolution of BAO we find that the second dip and third peak remains unaltered by non-linear effects. Furthermore, we find that the square of the damping factor and the shift of the position of BAO extrema scale to good approximation as the square of the growth factor, in agreement with expectations from perturbation theory. This confirms the idea that, besides cosmic distances, an accurate measurement of BAO at different redshifts can directly probe the growth of cosmic structures.

We have performed the first-ever numerical N- body simulation of the full observable universe (DEUS "Dark Energy Universe Simulation" FUR "Full Universe Run"). This has evolved 550 billion particles on an Adaptive Mesh Refinement grid with more than two trillion computing points along the entire evolutionary history of the universe and across 6 order of magnitudes length scales, from the size of the Milky Way to that of the whole observable universe. To date, this is the largest and most advanced cosmological simulation ever run. It provides unique information on the formation and evolution of the largest structure in the universe and an exceptional support to future observational programs dedicated to mapping the distribution of matter and galaxies in the universe. The simulation has run on 4752 (of 5040) thin nodes of BULL supercomputer CURIE, using more than 300 TB of memory for 10 million hours of computing time. About 50 PBytes of data were generated throughout the run. Using an advanced and innovative reduction workflow the amount of useful stored data has been reduced to 500 TBytes.