A substantial fraction of the stellar mass growth across cosmic time occurred within dust-enshrouded environments. Yet, the exact amount of star-forming activity that took place in high-redshift dusty galaxies currently missed by optical surveys has
been barely explored. Using the Spitzer observations of COSMOS we determined the fraction of luminous star-forming galaxies at 1.5<z<3 escaping the traditional color selection techniques because of dust extinction, as well as their contribution to the cosmic star formation density at high redshift. We find that the BzK criterion offers an almost complete (~90%) identification of the 24mic sources at 1.4<z<2.5, while the BM/BX criterion miss 50% of the MIPS population. Similarly the criterion based on the presence of a stellar bump in massive sources (so-called IRAC peakers) miss up to 40% of the IR luminosity density and only 25% of the IR energy density at z~2 is produced by Optically-Faint IR-bright galaxies selected based on their extreme mid-IR to optical flux ratios. We conclude that color selections of distant star-forming galaxies must be used with lots of care given the substantial bias they can suffer. In particular, the effect of dust extinction strongly impacts the completeness of identifications at the bright end of the bolometric luminosity function, which implies large and uncertain extrapolations to account for the contribution of dusty galaxies missed by these selections. In the context of forthcoming facilities that will operate at long wavelengths (e.g., $JWST$, ALMA, SAFARI, EVLA, SKA), this emphasizes the importance of minimizing the extinction biases when probing the activity of star formation in the early Universe.
We present the first results obtained from the identification of ~30,000 sources in the 24mic observations of the COSMOS field at S24>~80muJy. Using accurate photo-zs and extrapolations of the counts at faint fluxes we resolve with unprecedented deta
il the build-up of the mid-IR background across cosmic ages. We find that ~50% and ~80% of the 24mic background intensity originate from galaxies at z<~1 and z<~2 respectively, supporting the scenario where highly obscured sources at very high redshifts (z>~2) contribute only marginally to the CIB. Assuming flux-limited selections at optical wavelengths, we also find that the fraction of i-band sources with 24mic detection strongly increases up to z~2 as a consequence of the rapid evolution that star-forming galaxies have undergone with lookback time. Nonetheless this rising trend shows a clear break at z~1.3, probably due to k-correction effects implied by the complexity of SEDs in the mid-IR. Finally, we compare our results with the predictions from different models of galaxy formation. We note that semi-analytical formalisms fail to reproduce the redshift distributions observed at 24mic. Furthermore the simulated galaxies exhibit R-K colors much bluer than observed and the predicted K-band fluxes are systematically underestimated at z>0.5. Unless these discrepancies result from an incorrect treatment of extinction they may reflect an underestimate of the predicted density of high-z massive sources with strong star formation, which would point to more fundamental processes that are still not fully controlled in the simulations. The most recent backward evolution scenarios reproduce reasonably well the flux/redshift distribution of 24mic sources up to z~3, although none of them is able to exactly match our results at all redshifts. [Abridged]
