On the recovery of galaxy properties from SED fitting solutions

TL;DR

This study evaluates four inverse population synthesis codes' ability to accurately recover galaxy properties from spectral data, highlighting differences in bias and precision in their results.

Contribution

The paper compares four spectral fitting methods using a large synthetic galaxy dataset, revealing their relative accuracy and biases in recovering galaxy parameters.

Findings

DynBaS shows minimal systematic biases.

All methods have similar precision levels.

Spectral fits are almost perfect despite parameter differences.

Abstract

We explore the ability of four different inverse population synthesis codes to recover the physical properties of galaxies from their spectra by SED fitting. Three codes, DynBaS, TGASPEX, and GASPEX, have been implemented by the authors and are described in detail in the paper. STARLIGHT, the fourth code, is publicly available. DynBaS selects dynamically a different spectral basis to expand the spectrum of each target galaxy; TGASPEX uses an unconstrained age basis, whereas GASPEX and STARLIGHT use for all fits a fixed spectral basis selected a priori by the code developers. Variable and unconstrained basis reflect the peculiarities of the fitted spectrum and allow for simple and robust solutions to the problem of extracting galaxy parameters from spectral fits. We assemble a Synthetic Spectral Atlas of Galaxies (SSAG), comprising 100,000 galaxy spectra corresponding to an equal number of star formation histories based on the recipe of Chen et al. (2012). We select a subset of 120 galaxies from SSAG with a colour distribution similar to that of local galaxies in the seventh data release (DR7) of the Sloan Digital Sky Survey (SDSS) and produce 30 random noise realisations for each of these spectra. For each spectrum we recover the mass, mean age, metallicity, internal dust extinction, and velocity dispersion characterizing the dominant stellar population in the problem galaxy. All methods produce almost perfect fits to the target spectrum, but the recovered physical parameters can differ significantly. Our tests provide a quantitative measure of the accuracy and precision with which these parameters are recovered by each method. From a statistical point of view all methods yield similar precisions, whereas DynBaS produces solutions with minimal systematic biases in the distributions of residuals for all of these parameters.