We have analysed a large data set of OVI absorber candidates found in the spectra of 3702 SDSS quasars, focusing on a subsample of 387 AGN sightlines with an average S/N>5.0, allowing for detection of absorbers above rest-frame equivalents widths W_r>0.19 A for the OVI 1032 A component. Accounting for random interlopers mimicking an OVI doublet, we derive for the first time a secure lower limit for the redshift number density $Delta N / Delta z$ for redshifts z_abs>2.8. With extensive Monte Carlo simulations we quantify the losses of absorbers due to blending with the ubiquitous Lyman forest lines, and estimate the success rate of retrieving each individual candidate as a function of its redshift, the emission redshift of the quasar, the strength of the absorber and the measured S/N of the spectrum by modelling typical Ly forest spectra. These correction factors allow us to derive the incompleteness and S/N corrected redshift number densities of OVI absorbers :$Delta N _{OVI, c} / Delta z_{c} (2.8 < z < 3.2) = 4.6+-0.3, at 3.2 < z < 3.6 = 6.7+-0.8,and at 3.6 < z < 4.0 = 8.4+-2.9. We can place a secure lower limit for the contribution of OVI to the closure mass density at the redshifts probed here: $Omega _{OVI} (2.8 < z < 3.2) >1.9x10^{-8} h^-1. We show that the strong lines we probe account for over 65% of the mass in the OVI absorbers; the weak absorbers, while dominant in line number density, do not contribute significantly to the mass density. Making a conservative assumption about the ionisation fraction, and adopting the Anders (1989) solar abundance values, we derive the mean metallicty of the gas probed in our search : $zeta (2.8 < z < 3.2) > 3.6 x 10^-4 h, in good agreement with other studies. These results demonstrate that large spectroscopic datasets such as SDSS can play an important role in QSO absorption line studies, in spite of the relatively low resolution.
Download