Relation between dark matter halo mass and UV luminosity/SFR (left  and right axes) for all redshift samples. The bottom panels correspond  to the standard IC fitting procedure and the top panels to the fixed IC  fitting procedure. The panels on the left are the observed relations,  and the panels on the right are the results after corrections for dust  attenuation have been made. The solid lines correspond to the  predictions from the Millennium Simulation galaxy models (De Lucia &  Blaizot 2007), reduced to median relations.

We investigate  the luminosity dependent clustering of rest-frame UV selected galaxies  at z~4, 3, 2.2, and 1.7 in the Keck Deep Fields (KDFs), which are  complete to R = 27 and cover 169 arcmin^2. We find that at z~4 and 3,  UV-bright galaxies cluster more strongly than UV-faint ones, but at  z~2.2 and 1.7, the UV-bright galaxies are no longer the most strongly  clustered. We derive mass estimates for objects in our sample by  comparing our measurements to the predicted clustering of dark matter  haloes in the Millennium Simulation. From these estimates, we infer  relationships between halo mass and star formation rate (SFR), and find  that the most massive dark matter haloes in our sample host galaxies  with high SFRs (M_1700 < -20, or > 50 M_sun/yr) at z~3 and 4,  moderate SFRs (-20<M_1700<-19, or ~20 M_sun/yr) at z~2.2, and  lower SFRs (-19<M_1700<-18, or ~2 M_sun/yr) at z~1.7. We believe  our measurements may provide a new line of evidence for galaxy  downsizing by extending that concept from stellar to halo mass. We also  find that the objects with blue UV colors in our sample are much more  strongly clustered than those with red UV colors, and we propose that  this may be due to the presence of the 2175A dust absorption bump in  more massive halos, which contain the older stellar populations and dust  needed to produce the feature. The relatively small area covered by the  survey means that the absolute values of the correlation lengths and  halo masses we derive are heavily dependent on the "integral constraint"  correction, but the uniformly deep coverage across a large redshift  interval allows us to detect several important trends that are  independent of this correction.

See the websiade for more details.http://arxiv.org/abs/1106.2146 (SY)