Spectral synthesis is largely used in the literature to decompose stellar populations with integrated light of galaxies as if the star formation histories (SFH) could be approximated by single bursts. In the case of our method (see http://www.starlight.ufsc.br/ for the SEAGal - Semi Empirical Analysis of Galaxies - collaboration), the starlight code combines the spectra of simple stellar populations (SSP) of different ages and metallicities, computed with high spectral resolution evolutionary synthesis models of Bruzual & Charlot (2003), to reproduce the observed spectrum of a given galaxy from which we can derived a huge amount of galaxy properties such as: the population vector, stellar mass, extinction and others. We have done that for all galaxies of the SDSS database. Despite all the results of astrophysical interest, we have decided to use continuous composite stellar models (CSP) with a single metallicity and a star formation rate ∝ τ−1e−t/τ, where t stands for the time that the star formation started (1, 5 and 13 Gyr ago) and τ is the attenuation factor chosen to be 1, 5, 10 and 99 Gyr. When the attenuation with respect to the time t is very low, this mimics a single burst, and when we choose it to be very large (99 Gyr), this is almost a constant star formation rate. We have perturbed each composite model spectrum 10 times with three distinct signal/noise ratios equal to 10, 15 and 30 in λ0 = 4020 Å. These models were inserted into our code to verify how a picture of single bursts deal with continuous composite models of galaxies. Our CSP models can be easily integrated in an analytical form. Therefore, we have derived theoretically the mean ages and metallicities and compared them to the output derived by the synthesis. We can see that the synthesized mean ages weighted by light tend to be lower than the models, due to the degeneracies involved in the problem. The same thing can be found for the mean metallicities weighted by light, which tend to be higher for the output values.

Spectral synthesis of stellar populations has proven to be one of the most powerful methods to decompose the different mixtures of stellar contributions in galaxies, and applications of this technique routinely appear in the literature nowadays. Our group, for instance, the SEAGal (Semi Empirical Analysis of Galaxies) collaboration, has derived the star formation history of all galaxies in the SDSS with the starlight code, obtaining various results of astrophysical interest. As any other fossil method, the results rely heavily on high spectral resolution evolutionary synthesis models. To test this model dependence we run starlight on samples of star-forming and passive galaxies from the SDSS using different sets of models.

We explore models using “Padova 1994” and modified “Padova” evolutionary tracks with a different receipt for the asymptotic giant branch phase, as well as different stellar libraries (STELIB versus MILES+Granada). We then compare derived properties such as mean age, mean metallicity, extinction, star-formation and chemical histories. Despite a broad brush agreement, systematic differences emerge from this comparison. The different evolutionary tracks used lead to essentially the same results, at least insofar as optical spectra are concerned. Different stellar libraries, on the other hand, have a much bigger impact. The newer models produce quantifiably better fits and eliminate some pathologies (like suspicious combinations of base elements, systematical spectral residuals in some windows, and, sometimes, negative extinction) of fits derived with STELIB-based models, but there are still some caveats. These empirical tests provide useful feedback for model makers.