Detecting episodes of star formation using Bayesian model selection

Bayesian model comparison frameworks can be used when fitting models to data in order infer the appropriate complexity a data-driven manner. We aim use them detect correct number of major episodes star formation from analysis spectral energy distributions (SEDs) galaxies, modeled after 3D-HST galaxies at z ~ 1. Starting published stellar population properties these we kernel density estimates build multivariate input parameter obtain realistic simulations. create simulated sets spectra varying degrees (identified by parameters), and derive SED results evidences for pairs nested models, including as well more simplistic ones, using BAGPIPES codebase with sampling algorithm MultiNest. then ask question: is it true - expected that has larger evidence?} Our indicate ratio (the Bayes factor) able identify underlying vast majority cases. The quality improves primarily function total S/N SED. also compare factors obtained evidence those via Savage-Dickey Density Ratio (SDDR), an analytic approximation which calculated samples regular Markov Chain Monte Carlo methods. show SDDR satisfactorily replace full calculation provided sufficient.