The Incidence of Low-Metallicity Lyman-Limit Systems at z~3.5: Implications for the Cold-Flow Hypothesis of Baryonic Accretion

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 IGM. Here we report on an abundance survey of 17 H I-selected LLSs at $z=3.2$-4.4 which exhibit no metal absorption in SDSS 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 $N_{\rm HI}$ measurements typical of LLSs. The metal-poor LLS sample overlaps with the IGM in metallicity and is best described by a model where $71^{+13}_{-11}\%$ 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 $z\sim3.7$ could trace unprocessed gas. An ancillary sample of ten LLSs without any a priori metal-line selection is best fit with $48^{+14}_{-12}\%$ of metallicities drawn from the IGM. We compare these results with regions of a moving-mesh simulation; the simulation finds only half as many baryons in IGM-metallicity LLSs, and most of these lie beyond the virial radius of the nearest galaxy halo. 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.