Comparing halo bias from abundance and clustering

We model the abundance of haloes in the $\sim(3 \ \text{Gpc}/h)^3$ volume of the MICE Grand Challenge simulation by fitting the universal mass function with an improved Jack-Knife error covariance estimator that matches theory predictions. We present unifying relations between different fitting models and new predictions for linear ($b_1$) and non-linear ($c_2$ and $c_3$) halo clustering bias. Different mass function fits show strong variations in their performance when including the low mass range ($M_h \lesssim 3 \ 10^{12} \ M_{\odot}/h$) in the analysis. Together with fits from the literature we find an overall variation in the amplitudes of around $10$% in the low mass and up to $50$% in the high mass (galaxy cluster) range ($M_h > 10^{14} \ M_{\odot}/h$). These variations propagate into a $10$% change in $b_1$ predictions and a $50$% change in $c_2$ or $c_3$. Despite these strong variations we find universal relations between $b_1$ and $c_2$ or $c_3$ for which we provide simple fits. Excluding low mass haloes, different models fitted with reasonable goodness in this analysis, show percent level agreement in their $b_1$ predictions, but are systematically $5-10$% lower than the bias directly measured with two-point halo-mass clustering. This result confirms previous findings derived from smaller volumes (and smaller masses). Inaccuracies in the bias predictions lead to $5-10$% errors in growth measurements. They also affect any HOD fitting or (cluster) mass calibration from clustering measurements.