|Annu. Rev. Astron. Astrophys. 2013. 51:393-455
Copyright © 2013 by Annual Reviews. All rights reserved
The goal of this review has been to summarize what we have learned about galaxies from their panchromatic SEDs using the tools of stellar population synthesis. By way of concluding, I would like to turn toward the future and highlight areas where additional work is needed to make SED modeling both a more firmly quantitative science and also an effective engine for new discoveries. In short, and not surprisingly, the future requires better data, better models, better comparison of models to data, and a better understanding of what is in principle knowable from the analysis of galaxy SEDs.
A clear theme of this review has been the power of combining broadband data with moderate resolution spectra. The SDSS has been a revolutionizing force in this regard, as it has provided high-quality photometry and optical spectra for over one million objects. The great drawback of the SDSS is that it is fiber-based and thus the spectra only sample the central 3" (in diameter) of galaxies. This drawback will be alleviated with IFU spectroscopic surveys of nearby galaxies, including the recently completed SAURON survey (Bacon et al. 2001), the ongoing ATLAS3D survey (Cappellari et al. 2011) of 260 early-type galaxies, the CALIFA survey of ~ 600 galaxies (Sánchez et al. 2012), and the proposed MaNGA survey of a mass-limited sample of ~ 10,000 galaxies. Such surveys will provide high-quality spectra that are well-matched to the broadband data. Restframe NIR spectra will be a new frontier for SPS studies as next generation NIR facilities become operational, including ground-based spectrographs (FIRE, FMOS, KMOS, MOSFIRE, FLAMINGOS-2, etc.), and the eventual launch of the James Webb Space Telescope. High-quality models are only now being developed to interpret such data. In addition to spectra, narrow-band photometry will also be a valuable addition to the landscape, as pioneered by the COMBO-17 survey (Wolf et al. 2004) and now utilized by surveys such as ALHAMBRA (Moles et al. 2008), the NEWFIRM Medium-Band Survey (Whitaker et al. 2011), and SHARDS (Perez-Gonzalez et al. 2012). Grism data will also help bridge the gap between broadband photometry and moderate resolution spectra, as demonstrated by the 3D-HST survey (Brammer et al. 2012). In addition to these object-by-object surveys, the construction of composite SEDs from galaxies spanning a range of redshifts allows for the creation of very high-quality and well-sampled SEDs that will be invaluable for SPS studies (Assef et al. 2008, Kriek et al. 2011).
Another theme of this review has been the growing realization that uncertainties in the SPS models are becoming a critical limiting factor to the interpretation of galaxy SEDs. The challenge here is not simply to enumerate the uncertainties but rather to identify areas where clear progress can be made. As a first step, all SPS models should include contributions from nebular emission and dust around AGB stars, as these processes are known to occur and the incorporation of such effects into the models is reasonably straightforward, even if the details are uncertain. Panchromatic models (i.e., FUV-FIR coverage) should also become standard both because IR data are now widely available and because sophisticated dust emission models are well-developed (e.g., Draine & Li 2007). The stellar atmospheric and synthetic spectral models will benefit from new asteroseismology measurements from the Kepler mission, interferometric observations (e.g., by CHARA), and new very high resolution UV-NIR spectral atlases of nearby stars across the HR diagram (Bagnulo et al. 2003, Lebzelter et al. 2012). Perhaps the most vexing issues lie with the stellar evolution uncertainties, as obvious calibrating data are lacking. The well-known problem is that, while globular clusters are the canonical testing ground for stellar evolution, metal-rich clusters are rare, and so the models tend to be poorly constrained precisely in the metallicity range most relevant for modeling galaxies. Moreover, the largest uncertainties are associated with fast evolutionary phases and so stars in such phases will be rare in all but the most massive star clusters. Efforts to constrain uncertain stellar evolutionary phases from the SEDs of galaxies is promising because the right metallicity ranges are probed and there are sufficient numbers of stars to overcome Poisson noise, but the obvious complexity of dealing with composite stellar populations makes this approach challenging (e.g., MacArthur et al. 2010, Kriek et al. 2010, Zibetti et al. 2012). The Panchromatic Hubble Andromeda Treasury survey (Dalcanton et al. 2012) is an HST program covering ~ 1/3 of M31's star-forming disk in six filters. It promises to provide new and powerful constraints on luminous and advanced stellar evolutionary phases at moderately high metallicities.
More accurate models and higher-quality data will necessitate a more sophisticated approach to comparing the two. Presently, model fitting is something of an art, owing to the fact that large regions of parameter space are often severely under-constrained, which implies that the choice of priors on the model parameters can have a large impact on the derived results. A few basic guidelines should be followed to ensure that results are robust. For example, one should not simply fix a parameter to a particular value because it is under-constrained. Moreover, because the likelihood surface often contains multiple peaks and valleys and is frequently computed on a coarse grid, the best-fit parameters ought not be chosen based on the minimum of χ2 (see e.g., Taylor et al. 2011). Rather, the full posterior distributions should be used to derive best-fit values and associated uncertainties. The choice of priors needs to be considered carefully, and in fact the model space should probably depend on the type of data being fit, the redshift of the object, and even its spectral type (quiescent vs. star-forming vs. peculiar). As the number of parameters increases, Markov Chain Monte Carlo techniques will see more widespread use owing to their efficient exploration of parameter space. With regards to the modeling of moderate resolution spectra, the analysis of spectral indices should eventually give way to full spectral fitting as the latter not only allows for the extraction of more information but also allows the modeler to visually inspect the fits in a way that is not possible when only EWs are extracted from the data. This is important for identifying model systematics and areas for future improvement.
Finally, further work is needed to understand what is knowable, in principle, from the modeling of galaxy SEDs. Questions such as `how many discrete SF episodes can be measured from high quality optical spectra?', or `how many moments of the metallicity distribution function can be extracted from SEDs?', or `can the detailed dust attenuation curve be measured with sufficiently high quality data on an object-by-object basis?' have yet to be thoroughly explored. Addressing these questions will be difficult because they depend sensitively on the quality of the data, the SED type, and the reliability of the models. Fitting routines such as STARLIGHT, STECMAP, and VESPA that attempt a non-parametric recovery of the SFH and metallicity distribution function offer probably the most reliable tools to explore these questions. Theoretical studies aimed at understanding `what is knowable' will help guide the next generation of surveys aimed at studying the detailed physical properties of galaxies.
The author is not aware of any affiliations, memberships, funding, or financial holdings that might be perceived as affecting the objectivity of this review.
I would like to thank my collaborators for the continuous lively conversations that have helped form my views on this topic. I would also like to thank the authors who generously shared their figures for this review, and especially Elisabete da Cunha, John Moustakas, Naveen Reddy and Rita Tojeiro for providing new or modified figures. Stéphane Charlot, Daniel Dale, Sandy Faber, Jerome Fang, Ricardo Schiavon, Rita Tojeiro and Scott Trager are thanked for very useful comments on an earlier version of this manuscript.