The VIMOS Public Extragalactic Redshift Survey (VIPERS): An unbiased estimate of the growth rate of structure at $\mathbf{\left<z\right>=0.85}$ using the clustering of luminous blue galaxies

We use the VIPERS final data release to investigate the performance of colour-selected populations of galaxies as tracers of linear large-scale motions. We empirically select volume-limited samples of blue and red galaxies as to minimise the systematic error on the estimate of the growth rate $f\sigma_8$ from the anisotropy of the two-point correlation function. To this end, rather than rigidly splitting the sample into two colour classes we define the red/blue fractional contribution of each object through a weight based on the $(U-V)$ colour distribution. Using mock surveys that are designed to reproduce the observed properties of VIPERS galaxies, we find the systematic error in recovering the fiducial value of $f\sigma_8$ to be minimized when using a volume-limited sample of luminous blue galaxies. We model non-linear corrections via the Scoccimarro extension of the Kaiser model, finding systematic errors on $f\sigma_8$ of below $1-2\%$, using scales as small as 5 $h^{-1}\mathrm{Mpc}$. We interpret this result as indicating that selection of luminous blue galaxies maximises the fraction that are central objects in their dark matter haloes; this in turn minimises the contribution to the measured $\xi(r_p,\pi)$ from the 1-halo term, which is dominated by non-linear motions. The gain is inferior if one uses the full magnitude-limited sample of blue objects, consistent with the presence of a significant fraction of blue, fainter satellites dominated by non-streaming, orbital velocities. We measure a value of $f\sigma_8=0.45 \pm 0.11$ over the single redshift range $0.6\le z\le 1.0$, corresponding to an effective redshift for the blue galaxies $\left<z\right>=0.85$. Including in the likelihood the potential extra information contained in the blue-red galaxy cross-correlation function does not lead to an appreciable improvement in the error bars, while it increases the systematic error.