Bayesian model averaging is a procedure to obtain parameter constraints that account for the uncertainty about the correct cosmological model. We use recent cosmological observations and Bayesian model averaging to derive tight limits on the curvature parameter, as well as robust lower bounds on the curvature radius of the Universe and its minimum size, while allowing for the possibility of an evolving dark energy component. Because flat models are favoured by Bayesian model selection, we find that model-averaged constraints on the curvature and size of the Universe can be considerably stronger than non model-averaged ones. For the most conservative prior choice (based on inflationary considerations), our procedure improves on non model-averaged constraints on the curvature by a factor of ~ 2. The curvature scale of the Universe is conservatively constrained to be R_c > 42 Gpc (99%), corresponding to a lower limit to the number of Hubble spheres in the Universe N_U > 251 (99%).

We extend our study of the optimization of large baryon acoustic oscillation (BAO) surveys to return the best constraints on the dark energy, building on Paper I of this series (Parkinson et al. 2007). The survey galaxies are assumed to be pre-selected active, star-forming galaxies observed by their line emission with a constant number density across the redshift bin. Star-forming galaxies have a redshift desert in the region 1.6 < z < 2, and so this redshift range was excluded from the analysis. We use the Seo & Eisenstein (2007) fitting formula for the accuracies of the BAO measurements, using only the information for the oscillatory part of the power spectrum as distance and expansion rate rulers. We go beyond our earlier analysis by examining the effect of including curvature on the optimal survey configuration and updating the expected `prior' constraints from Planck and SDSS. We once again find that the optimal survey strategy involves minimizing the exposure time and maximizing the survey area (within the instrumental constraints), and that all time should be spent observing in the low-redshift range (z<1.6) rather than beyond the redshift desert, z>2. We find that when assuming a flat universe the optimal survey makes measurements in the redshift range 0.1 < z <0.7, but that including curvature as a nuisance parameter requires us to push the maximum redshift to 1.35, to remove the degeneracy between curvature and evolving dark energy. The inclusion of expected other data sets (such as WiggleZ, BOSS and a stage III SN-Ia survey) removes the necessity of measurements below redshift 0.9, and pushes the maximum redshift up to 1.5. We discuss considerations in determining the best survey strategy in light of uncertainty in the true underlying cosmological model.

The question of determining the spatial geometry of the Universe is of greater relevance than ever, as precision cosmology promises to verify inflationary predictions about the curvature of the Universe. We revisit the question of what can be learnt about the spatial geometry of the Universe from the perspective of a three-way Bayesian model comparison. We show that, given current data, the probability that the Universe is spatially infinite lies between 67% and 98%, depending on the choice of priors. For the strongest prior choice, we find odds of order 50:1 (200:1) in favour of a flat Universe when compared with a closed (open) model. We also report a robust, prior-independent lower limit to the number of Hubble spheres in the Universe, N_U > 5 (at 99% confidence). We forecast the accuracy with which future CMB and BAO observations will be able to constrain curvature, finding that a cosmic variance limited CMB experiment together with an SKA-like BAO observation will constrain curvature with a precision of about sigma ~ 4.5x10^{-4}. We demonstrate that the risk of 'model confusion' (i.e., wrongly favouring a flat Universe in the presence of curvature) is much larger than might be assumed from parameter errors forecasts for future probes. We argue that a 5-sigma detection threshold guarantees a confusion- and ambiguity-free model selection. Together with inflationary arguments, this implies that the geometry of the Universe is not knowable if the value of the curvature parameter is below |Omega_curvature| ~ 10^{-4}, a bound one order of magnitude larger than the size of curvature perturbations, ~ 10^{-5}. [abridged]