Subhalo abundance matching and assembly bias in the EAGLE simulation

Subhalo abundance matching (SHAM) is a widely-used method to connect galaxies with dark matter structures in numerical simulations. SHAM predictions agree remarkably well with observations, yet they still lack strong theoretical support. We examine the performance, implementation, and assumptions of SHAM using the EAGLE project simulations. We find that $V_{\rm relax}$, the highest value of the circular velocity attained by a subhalo while it satisfies a relaxation criterion, is the subhalo property that correlates most strongly with galaxy stellar mass ($M_{\rm star}$). Using this parameter in SHAM, we retrieve the real-space clustering of EAGLE to within our statistical uncertainties on scales greater than $2$ Mpc for galaxies with $8.77<\log_{10}(M_{\rm star}[M_\odot])<10.77$. Conversely, clustering is overestimated by $30\%$ on scales below $2$ Mpc for galaxies with $8.77<\log_{10}(M_{\rm star}[M_\odot])<9.77$ because SHAM slightly overpredicts the fraction of satellites in massive haloes compared to EAGLE. The agreement is even better in redshift-space, where the clustering is recovered to within our statistical uncertainties for all masses and separations. Additionally, we analyse the dependence of galaxy clustering on properties other than halo mass, i.e. the assembly bias. We demonstrate assembly bias alters the clustering in EAGLE by $20\%$ and that $V_{\rm relax}$ captures its effect to within $15\%$. We trace small differences in the clustering to the failure of SHAM as typically implemented, i.e. the $M_{\rm star}$ assigned to a subhalo does not depend on i) its host halo mass, ii) whether it is a central or a satellite. In EAGLE we find that these assumptions are not completely satisfied.