Non-parametric Star Formation History Reconstruction with Gaussian Processes I: Counting Major Episodes of Star Formation

Abstract: The star formation histories (SFHs) of galaxies contain imprints of the physical processes responsible for regulating star formation during galaxy growth and quenching. We improve the Dense Basis SFH reconstruction method of Iyer & Gawiser (2017), introducing a nonparametric description of the SFH based on the lookback times at which a galaxy assembles certain quantiles of its stellar mass. The method uses Gaussian Processes to create smooth SFHs that are independent of any functional form, with a flexible number of parameters that is adjusted to extract the maximum possible amount of SFH information from the SEDs being fit. We apply the method to reconstruct the SFHs of 48,791 galaxies with $H<25$ at $0.5 < z < 3.0$ across the five CANDELS fields. Using these SFHs, we study the evolution of galaxies as they grow more massive over cosmic time. We quantify the fraction of galaxies that show multiple major episodes of star formation, finding that the median time between two peaks of star formation is $\sim 0.42_{-0.10}^{+0.15}t_{univ}$ Gyr, where $t_{univ}$ is the age of the universe at a given redshift and remains roughly constant with stellar mass. Correlating SFHs with morphology, we find that studying the median SFHs of galaxies at $0.5<z<1.0$ at the same mass ($10^{10}< M_* < 10^{10.5}M_\odot$) allows us to compare the timescales on which the SFHs decline for different morphological classifications, ranging from $0.60^{-0.54}_{+1.54}$ Gyr for galaxies with spiral arms to $2.50^{-1.50}_{+2.25}$ Gyr for spheroids. The Gaussian Process-based SFH description provides a general approach to reconstruct smooth, nonparametric SFH posteriors for galaxies with a flexible number of parameters that can be incorporated into Bayesian SED fitting codes to minimize the bias in estimating physical parameters due to SFH parametrization.