We explore star formation histories (SFHs) of galaxies based on the evolution of the star formation rate stellar mass relation (SFR-M * ). Using data from the FourStar Galaxy Evolution Survey (ZFOURGE) in combination with far-IR imaging from the Spitzer and Herschel observatories we measure the SFR-M * relation at 0.5 < z < 4. Similar to recent works we find that the average infrared spectral energy distributions of galaxies are roughly consistent with a single infrared template across a broad range of redshifts and stellar masses, with evidence for only weak deviations. We find that the SFR-M * relation is not consistent with a single power law of the form M SFR * µ a at any redshift; it has a power law slope of α ∼ 1 at low masses, and becomes shallower above a turnover mass (M 0 ) that ranges from 10 9.5 to 10 10.8 M e , with evidence that M 0 increases with redshift. We compare our measurements to results from state-of-the-art cosmological simulations, and find general agreement in the slope of the SFR-M * relation albeit with systematic offsets. We use the evolving SFR-M * sequence to generate SFHs, finding that typical SFRs of individual galaxies rise at early times and decline after reaching a peak. This peak occurs earlier for more massive galaxies. We integrate these SFHs to generate mass growth histories and compare to the implied mass growth from the evolution of the stellar mass function (SMF). We find that these two estimates are in broad qualitative agreement, but that there is room for improvement at a more detailed level. At early times the SFHs suggest mass growth rates that are as much as 10× higher than inferred from the SMF. However, at later times the SFHs under-predict the inferred evolution, as is expected in the case of additional growth due to mergers.

Using observations from the FourStar Galaxy Evolution Survey (ZFOURGE), we obtain the deepest measurements to date of the galaxy stellar mass function at 0.2 < z < 3. ZFOURGE provides wellconstrained photometric redshifts made possible through deep medium-bandwidth imaging at 1-2µm. We combine this with HST imaging from the Cosmic Assembly Near-IR Deep Extragalactic Legacy Survey (CANDELS), allowing for the efficient selection of both blue and red galaxies down to stellar masses ∼ 10 9.5 M ⊙ at z ∼ 2.5. The total surveyed area is 316 arcmin 2 distributed over three independent fields. We supplement these data with the wider and shallower NEWFIRM Medium-Band Survey (NMBS) to provide stronger constraints at high masses. Several studies at z ≤ 1.5 have revealed a steepening of the slope at the low-mass end of the stellar mass function (SMF), leading to an upturn at masses < 10 10 M ⊙ that is not well-described by a standard single-Schechter function. We find evidence that this feature extends to at least z ∼ 2, and that it can be found in both the star-forming and quiescent populations individually. The characteristic mass (M * ) and slope at the lowest masses (α) of a double-Schechter function fit to the SMF stay roughly constant at Log(M/M ⊙ ) ∼ 10.65 and ∼ −1.5 respectively. The SMF of star-forming galaxies has evolved primarily in normalization, while the change in shape is relatively minor. Our data allow us for the first time to observe a rapid buildup at the low-mass end of the quiescent SMF. Since z = 2.5, the total stellar mass density of quiescent galaxies (down to 10 9 M ⊙ ) has increased by a factor of ∼12 whereas the mass density of star-forming galaxies only increases by a factor of ∼2.2. * This paper includes data gathered with the 6.5 meter Magellan Telescopes located at Las Campanas Observatory, Chile.

We report the likely identification of a substantial population of massive M ∼ 10 11 M galaxies at z ∼ 4 with suppressed star formation rates (SFRs), selected on rest-frame optical to near-IR colors from the FourStar Galaxy Evolution Survey (ZFOURGE). The observed spectral energy distributions show pronounced breaks, sampled by a set of near-IR medium-bandwidth filters, resulting in tightly constrained photometric redshifts. Fitting stellar population models suggests large Balmer/4000 Å breaks, relatively old stellar populations, large stellar masses, and low SFRs, with a median specific SFR of 2.9 ± 1.8 × 10 −11 yr −1 . Ultradeep Herschel/PACS 100 μm, 160 μm and Spitzer/MIPS 24 μm data reveal no dust-obscured SFR activity for 15/19(79%) galaxies. Two far-IR detected galaxies are obscured QSOs. Stacking the far-IR undetected galaxies yields no detection, consistent with the spectral energy distribution fit, indicating independently that the average specific SFR is at least 10× smaller than that of typical star-forming galaxies at z ∼ 4. Assuming all far-IR undetected galaxies are indeed quiescent, the volume density is 1.8 ± 0.7 × 10 −5 Mpc −3 to a limit of log 10 M/M 10.6, which is 10× and 80× lower than at z = 2 and z = 0.1. They comprise a remarkably high fraction (∼35%) of z ∼ 4 massive galaxies, suggesting that suppression of star formation was efficient even at very high redshift. Given the average stellar age of 0.8 Gyr and stellar mass of 0.8 × 10 11 M , the galaxies likely started forming stars before z = 5, with SFRs well in excess of 100 M yr −1 , far exceeding that of similarly abundant UV-bright galaxies at z 4. This suggests that most of the star formation in the progenitors of quiescent z ∼ 4 galaxies was obscured by dust.