SEDflow: Accelerated Bayesian SED Modeling using Amortized Neural Posterior Estimation

SEDflow is an accelerated Bayesian SED modeling method that uses the  Hahn et al. (2022a) PROVABGS SED model and Amortized Neural Posterior Estimation (ANPE) to derive posterior probability distributions of galaxy properties from optical photometry. SEDflow is\(10^5\times\) faster than conventional Markov Chain Monte Carlo sampling methods and takes ~1 second per galaxy to obtain posteriors. This repository includes all of the data used to train, validate, and test SEDflow.

This repository also includes a value-added catalog with detailed physical properties of 33,884 galaxies in the NASA-Sloan Atlas (http://www.nsatlas.org/). The properties are inferred from optical photometry in the u, g, r, i, z bands using SEDflow. For more details on this catalog and SEDflow see the documentation and Hahn & Melchior (2022). 

For each galaxy, the catalog provides posteriors of: 

• log_mstar: log10 of stellar mass
• log_sfr_1gyr: log10 of average star formation rate over 1Gyr
• log_z_mw: log10 of mass-weighted metallicity
• beta1, beta2, beta3, beta4: coefficients of the non-negative matrix factorization (NMF) star formation history basis functions
• fburst: fraction of stellar mass formed by a starburst event
• tburst: time of the starburst event
• log_gamma1, log_gamma2: log10 of coefficients of the NMF metallicity history basis functions
• tau_bc: birth cloud optical depth
• tau_ism: diffuse dust optical depth
• n_dust: Calzetti (2001) dust index