Ultraviolet to Mid-Infrared Observations of Star-forming Galaxies at z ~ 2: Stellar Masses and Stellar Populations

We present the broadband UV through mid-infrared spectral energy distributions (SEDs) of a sample of 72 spectroscopically confirmed star-forming galaxies at z = 2.30 ± 0.3. Located in a 72 arcmin2 field centered on the bright background QSO, HS 1700+643, these galaxies were preselected to lie at z ~ 2 solely on the basis of their rest-frame UV colors and luminosities and should be representative of UV-selected samples at high redshift. In addition to deep ground-based photometry spanning from 0.35 to 2.15 μm, we make use of Spitzer IRAC data, which probe the rest-frame near-IR at z ~ 2. The range of stellar populations present in the sample is investigated with simple, single-component stellar population synthesis models. The inability to constrain the form of the star formation history limits our ability to determine the parameters of extinction, age, and star formation rate without using external multiwavelength information. Emphasizing stellar mass estimates, which are much less affected by these uncertainties, we find ⟨log M*/M☉⟩ = 10.32 ± 0.51 for the sample. The addition of Spitzer IRAC data as a long-wavelength baseline reduces stellar mass uncertainties by a factor of 1.5-2 relative to estimates based on optical-Ks photometry alone. However, the total stellar mass estimated for the sample is remarkably insensitive to the inclusion of IRAC data. We find correlations between stellar mass and rest-frame R band (observed Ks) and rest-frame 1.4 μm (observed 4.5 μm) luminosities, although with significant scatter. Even at rest-frame 1.4 μm, the mass-to-light ratio varies by a factor of 15 indicating that even the rest-frame near-IR, when taken alone, is a poor indicator of stellar mass in star-forming galaxies at z ~ 2. Allowing for the possibility of episodic star formation, we find that typical galaxies in our sample could contain up to 3 times more stellar mass in an old underlying burst than what was inferred from single-component modeling. In contrast, mass estimates for the most massive galaxies in the sample (M* > 1011 M☉) are fairly insensitive to the method used to model the stellar population. Galaxies in this massive tail, which are also the oldest objects in the sample, could plausibly evolve into the passive galaxies discovered at z ~ 1.7 with near-IR selection techniques. In the general framework of hierarchical galaxy formation and mergers, which implies episodic star formation histories, galaxies at high redshift may pass in and out of UV-selected and near-IR color-selected samples as they evolve from phases of active star formation to quiescence and back again.