Detection of anisotropic galaxy assembly bias in BOSS DR12

We present evidence of anisotropic galaxy assembly bias in the Baryon Oscillation Spectroscopic Survey Data Release 12 galaxy sample at a level exceeding $5\sigma$. We use measurements of the line-of-sight velocity dispersion $\sigma_\star$ and stellar mass $M_\star$ to perform a simple split into subsamples of galaxies. We show that the amplitude of the monopole and quadrupole moments of the power spectrum depend differently on $\sigma_\star$ and $M_\star$, allowing us to split the galaxy sample into subsets with matching monopoles but significantly different quadrupoles on all scales. Combining data from the LOWZ and CMASS NGC galaxy samples, we find $>5\sigma$ evidence for anisotropic bias on scales $k<0.15\,h\,\rm{Mpc}^{-1}$. We also examine splits using other observed properties. For galaxy samples split using $M_\star$ and projected size $R_0$, we find no significant evidence of anisotropic bias. Galaxy samples selected using additional properties exhibit strongly varying degrees of anisotropic assembly bias, depending on which combination of properties is used to split into subsets. This may explain why previous searches for this effect using the Fundamental Plane found inconsistent results. We conclude that any selection of a galaxy sample that depends on $\sigma_\star$ can give biased and incorrect Redshift Space Distortion measurements.