ON STAR FORMATION RATES AND STAR FORMATION HISTORIES OF GALAXIES OUT TO z ~ 3.

We compare multi-wavelength star formation rate (SFR) indicators out to z ~ 3 in the GOODS-South field. Our analysis uniquely combines U to 8 µm photometry from FIREWORKS, MIPS 24 µm and PACS 70, 100, and 160 µm photometry from the PEP, and Hα spectroscopy from the SINS survey. We describe a set of conversions that lead to a continuity across SFR indicators. A luminosity-independent conversion from 24/um to total infrared luminosity yields estimates of L_IR that are in the median consistent with the L_IR derived from PACS photometry, albeit with significant scatter. Dust correction methods perform well at low-to-intermediate levels of star formation. They fail to recover the total amount of star formation in systems with large SFR_IR/SFR_UV ratios, typically occuring at the highest SFRs (SFR_UV+IR ≳ 100 M_⊙ yr_-1) and redshifts (z ≳ 2.5) probed. Finally, we confirm that Hα-based SFRs at 1.5 < z < 2.6 are consistent with SFR_SED and SFR_UV+IR provided extra attenuation toward H II regions is taken into account (A_V,neb = A_V,continuum/0.44). With the cross-calibrated SFR indicators in hand, we perform a consistency check on the star formation histories inferred from spectral energy distribution (SED) modeling. We compare the observed SFR-M relations and mass functions at a range of redshifts to equivalents that are computed by evolving lower redshift galaxies backward in time. We find evidence for underestimated stellar ages when no stringent constraints on formation epoch are applied in SED modeling. We demonstrate how resolved SED modeling, or alternatively deep UV data, may help to overcome this bias. The age bias is most severe for galaxies with young stellar populations and reduces toward older systems. Finally, our analysis suggests that SFHs typically vary on timescales that are long (at least several 100 Myr) compared to the galaxies' dynamical time. [ABSTRACT FROM AUTHOR]