Recent discussions in the literature about the importance of the contributions of very lumi- nous giant stars to the integrated near-IR light of galaxies (e.g., Maraston 2005; Maraston et al. 2006) highlighted how the use of different sets of stellar models can influence the results of stellar mass studies and conclusions about mass assembly at high redshift. Thus, we have chosen in this chapter to use two popular population synthesis models, those of BC03 and Maraston (2005), to model the SEDs of SMGs and estimate stellar masses. Here, we compare the fits to the SMG SEDs of the models of the different authors in an effort to determine if one set of models should be preferred over the other.
We plot the best-fit CSF models of BC03 (black line) and Maraston (2005) (green line) over the observed data points in Figures 6.3–6.10, noting that the fits of the burst models are similar so the CSF models are representative of the fits of the different star formation histories. For the most part, the fits of the BC03 and Maraston (2005) models appear very similar, though they almost always correspond to different ages and, thus, reddenings. In
the burst model fits, the ages agree to within 10 Myr for ∼50% of our SMG sample, but
differ by more than an order of magnitude only ∼10% of the time. When the ages agree
to within 10 Myr, the reddening values are usually within 0.3 mag of each other, and the reddenings inferred from the BC03 tend to be higher than those inferred from the Maraston (2005) model even when the BC03 age is older. In the CSF model fits, the ages agree to
within 1 Gyr for ∼ 60% of the sample and differ by more than an order of magnitude in
only∼10% of cases; similar to the burst models, the reddening values inferred are typically
within 0.3 of each other when the ages are similar, with BC03 reddenings tending to be larger. Thus, it seems that the BC03 models tend to require more reddening to match the SEDs of SMGs. However, as mentioned earlier, we do not find that one set of models gives systematically lower ages than the other, overall.
We note that both sets of models give ages older than the age of the universe for∼25%
of the sample using the CSF models, which in most cases can be attributed to the low time resolution of the age grids at ages above 2 Gyr. In computing average ages, in cases where the fitted age is greater than the age of the Universe at the redshift of the galaxy, we have
substituted the age of the Universe at that redshift in place of the fitted age.
Also worthy of note is that the models of the different authors fitted to the SMG SEDs differ most at the extreme blue and red ends of the SED. When there is a clear difference, the model of Maraston (2005) is usually brighter at the blue end of the SED and fainter at the red end. This behavior occurs when the BC03 model fit gives a higher reddening than the Maraston (2005) model fit, but the ages from the different models are similar, signaling that the BC03 models tend to be more blue than the Maraston (2005) models and, thus, require higher reddening to produce the observed ratio of near-IR to UV/optical light.
For the most part, the fits of both sets of models match the data reasonably well, even
though the formalχ2 values per degree of freedom of the fits are frequently greater than 1,
a result also found by Papovich et al. (2001) when fitting the optical-to-near-IR SEDs of
Lyman break galaxies (LBGs). The high values ofχ2imply that one or more of the following
are true: (1) the photometric errors on the measurements have been underestimated, (2) the distribution in the errors of the fitted parameters is not Gaussian, or (3) the models used do not adequately describe the SEDs of the SMGs. It is certainly likely that the stellar population models are insufficient to describe the observed SEDs, since, for example, we assume a single metallicity for all galaxies, we use only simple star formation histories, and we assume a single extinction model for galaxies which are likely to have a range of extinction characteristics. However, it is also possible that the photometric errors have been underestimated, especially in the IRAC data, because mosaic image pixels within a given region have correlated noise properties due to drizzling. To check if the photometric
errors were at the root of the high χ2 values, we re-fit the observed SEDs, including extra
5-15% systematic uncertainties in all of the photometric data. However, the changes in
the χ2 distributions of the fitted parameters were relatively insignificant when the extra
uncertainty was included, suggesting that it is more likely that the models are not good descriptions for all of the data.
For several objects (e.g., SMM J030244.82+000632.3), the best-fitting models from both BC03 and Maraston (2005) are clearly poor fits to the observed data, and as a result, the interpolated absolute magnitude that results from the fit is much lower than the observed data points would suggest. Since the absolute magnitudes in these cases may be in serious error, we exclude them from the sample averages of age and absolute magnitude.