We investigate the star formation histories (SFHs) of high redshift (3 <~ z <~ 5) star-forming galaxies selected based on their rest-frame ultraviolet (UV) colors in the CANDELS/GOODS-S field. By comparing the results from the spectral-energy-distrib
ution-fitting analysis with two different assumptions about the SFHs --- i.e., exponentially declining SFHs as well as increasing ones, we conclude that the SFHs of high-redshift star-forming galaxies increase with time rather than exponentially decline. We also examine the correlations between the star formation rates (SFRs) and the stellar masses. When the galaxies are fit with rising SFRs, we find that the trend seen in the data qualitatively matches the expectations from a semi-analytic model of galaxy formation. The mean specific SFR is shown to increase with redshift, also in agreement with the theoretical prediction. From the derived tight correlation between stellar masses and SFRs, we derive the mean SFH of star-forming galaxies in the redshift range of 3 <~ z <~ 5, which shows a steep power-law (with power alpha = 5.85) increase with time. We also investigate the formation timescales and the mean stellar population ages of these star-forming galaxies. Our analysis reveals that UV-selected star-forming galaxies have a broad range of the formation redshift. The derived stellar masses and the stellar population ages show positive correlation in a sense that more massive galaxies are on average older, but with significant scatter. This large scatter implies that the galaxies mass is not the only factor which affects the growth or star formation of high-redshift galaxies.
We explore methods to improve the estimates of star formation rates and mean stellar population ages from broadband photometry of high redshift star-forming galaxies. We use synthetic spectral templates with a variety of simple parametric star format
ion histories to fit broadband spectral energy distributions. These parametric models are used to infer ages, star formation rates and stellar masses for a mock data set drawn from a hierarchical semi-analytic model of galaxy evolution. Traditional parametric models generally assume an exponentially declining rate of star-formation after an initial instantaneous rise. Our results show that star formation histories with a much more gradual rise in the star formation rate are likely to be better templates, and are likely to give better overall estimates of the age distribution and star formation rate distribution of Lyman break galaxies. For B- and V-dropouts, we find the best simple parametric model to be one where the star formation rate increases linearly with time. The exponentially-declining model overpredicts the age by 100 % and 120 % for B- and V-dropouts, on average, while for a linearly-increasing model, the age is overpredicted by 9 % and 16 %, respectively. Similarly, the exponential model underpredicts star-formation rates by 56 % and 60 %, while the linearly-increasing model underpredicts by 15 % 22 %, respectively. For U-dropouts, the models where the star-formation rate has a peak (near z ~ 3) provide the best match for age -- overprediction is reduced from 110 % to 26 % -- and star-formation rate -- underprediction is reduced from 58 % to 22 %. We classify different types of star-formation histories in the semi-analytic models and show how the biases behave for the different classes. We also provide two-band calibration formulae for stellar mass and star formation rate estimations.
We investigate the biases and uncertainties in estimates of physical parameters of high-redshift Lyman break galaxies (LBGs), such as stellar mass, mean stellar population age, and star formation rate (SFR), obtained from broad-band photometry. By co
mbining LCDM hierarchical structure formation theory, semi-analytic treatments of baryonic physics, and stellar population synthesis models, we construct model galaxy catalogs from which we select LBGs at redshifts z ~ 3.4, 4.0, and 5.0. The broad-band spectral energy distributions (SEDs) of these model LBGs are then analysed by fitting galaxy template SEDs derived from stellar population synthesis models with smoothly declining SFRs. We compare the statistical properties of LBGs physical parameters -- such as stellar mass, SFR, and stellar population age -- as derived from the best-fit galaxy templates with the intrinsic values from the semi-analytic model. We find some trends in these distributions: first, when the redshift is known, SED-fitting methods reproduce the input distributions of LBGs stellar masses relatively well, with a minor tendency to underestimate the masses overall, but with substantial scatter. Second, there are large systematic biases in the distributions of best-fit SFRs and mean ages, in the sense that single-component SED-fitting methods underestimate SFRs and overestimate ages. We attribute these trends to the different star formation histories predicted by the semi-analytic models and assumed in the galaxy templates used in SED-fitting procedure, and to the fact that light from the current generation of star-formation can hide older generations of stars. These biases, which arise from the SED-fitting procedure, can significantly affect inferences about galaxy evolution from broadband photometry.
