2015ApJ...812...58C

Cold accretion is a primary growth mechanism of simulated galaxies, yet observational evidence of ''cold flows'' at redshifts where they should be most efficient (z = 2-4) is scarce. In simulations, cold streams manifest as Lyman-limit absorption systems (LLSs) with low heavy-element abundances similar to those of the diffuse intergalactic medium (IGM). Here we report on an abundance survey of 17 H i-selected LLSs at z = 3.2-4.4 which exhibits no metal absorption in Sloan Digital Sky Survey spectra. Using medium-resolution spectra obtained at Magellan, we derive ionization-corrected metallicities (or limits) with a Markov-chain Monte Carlo sampling that accounts for the large uncertainty in N_ H i_measurements typical of LLSs. The metal-poor LLS sample overlaps with the IGM in metallicity and can be described by a model where are drawn from the IGM chemical abundance distribution. These represent roughly half of all LLSs at these redshifts, suggesting that 28%-40% of the general LLS population at z ∼ 3.7 could trace accreting gas. An ancillary sample of ten LLSs without any a priori metal-line selection is fit by a model having of metallicities drawn from the IGM. We compare these results with regions of a moving-mesh simulation. The observed and simulated LLS metallicity distributions are in good agreement, after accounting for known uncertainties in both, with the fraction of simulated baryons in IGM-metallicity LLSs within a factor of two of the observed value. A statistically significant fraction of all LLSs have low metallicity and therefore represent candidates for accreting gas; large-volume simulations can establish what fraction of these candidates actually lie near galaxies and the observational prospects for detecting the presumed hosts in emission.