Cosmological parameters from a million photometric redshifts of SDSS Luminous Red Galaxies

We analyze MegaZ-LRG, a new photometric-redshift catalogue of Luminous Red Galaxies (LRGs) based on the imaging data of the Sloan Digital Sky Survey (SDSS) 4th Data Release. MegaZ-LRG, presented in a companion paper, contains > 10 photometric redshifts derived with ANNz, an Artificial Neural Network method, constrained by a spectroscopic sub-sample of ≈ 13,000 galaxies obtained by the 2dF-SDSS LRG and Quasar (2SLAQ) survey. The catalogue spans the redshift range 0.4 < z < 0.7 with an r.m.s. redshift error δz ≈ 0.03(1 + z). We measure the large-scale structure using two methods: (1) a spherical harmonic analysis in redshift slices, and (2) a direct re-construction of the spatial clustering pattern using Fourier techniques. We present the first cosmological parameter fits to galaxy angular power spectra from a photometric redshift survey. Combining the redshift slices with appropriate covariances, we determine best-fitting values for the matter density Ωm and baryon density Ωb which we express in the combinations Ωmh = 0.20 ± 0.03 and Ωb/Ωm = 0.14 ± 0.04 (with the Hubble parameter h and scalar index of primordial fluctuations held fixed). These results are in agreement with and independent of the latest studies of the Cosmic Microwave Background radiation, and their precision is comparable to analyses of contemporary spectroscopic-redshift surveys. We find visual suggestions of baryon oscillations in the clustering pattern, with a “model-independent” statistical significance of less than 3-σ. On the largest scales probed by the survey we measure an excess of power with respect to the model with a significance of ≈ 2-σ. We perform an extensive series of tests which conclude that our power spectrum measurements are robust against potential systematic photometric errors in the catalogue. We conclude that photometric-redshift surveys are competitive with spectroscopic surveys for measuring cosmological parameters in the simplest “vanilla” models. Future deep imaging surveys have great potential for further improvement, provided that systematic errors can be controlled.