Assembly bias in the local PNG halo bias and its implication for f_NL constraints

We use N-body simulations to study halo assembly bias (i.e., the dependence of halo clustering on properties beyond total mass) in the density and primordial non-Gaussianity (PNG) linear bias parameters b(1) and b(phi), respectively. We consider concentration, spin and sphericity as secondary halo properties, for which we find a clear detection of assembly bias for b(1) and b(phi). At fixed total mass, halo spin and sphericity impact b(1) and b(phi) in a similar manner, roughly preserving the shape of the linear b(phi)(b(1)) relation satisfied by the global halo population. Halo concentration, however, drives b(1) and b(phi) in opposite directions. This induces significant changes to the b(phi)(b(1)) relation, with higher concentration halos having higher amplitude of b(phi)(b(1)). For z = 0.5 and b(1) approximate to 2 in particular, the population comprising either all halos, those with the 33% lowest or those with the 33% highest concentrations have a PNG bias of b(phi) approximate to 3, b(phi) approximate to -1 and b(phi) approximate to 9, respectively. Varying the halo concentration can make b(phi) very small and even change its sign. These results have important ramifications for galaxy clustering constraints of the local PNG parameter f(NL) that assume fixed forms for the b(phi)(b(1)) relation. We illustrate the significant impact of halo assembly bias in actual data using the BOSS DR12 galaxy power spectrum: assuming that BOSS galaxies are representative of all halos, the 33% lowest or the 33% highest concentration halos yields sigma(fNL) = 44, 165, 19, respectively. Our results suggest taking host halo concentration into account in galaxy selection strategies to maximize the signal-to-noise on f(NL). They also motivate more simulation-based efforts to study the b(phi)(b(1)) relation of halos and galaxies.