Measuring the dust attenuation law of galaxies using photometric data

TL;DR

This study uses SED fitting with BAGPIPES to analyze galaxy dust attenuation laws, demonstrating that including infrared data helps accurately recover parameters and confirms that observed correlations are physical, not artifacts.

Contribution

The paper shows that infrared photometry improves the recovery of dust attenuation parameters and confirms the physical nature of the $A_V$ and $\delta$ correlation.

Findings

Infrared data reduces degeneracy between $A_V$ and star formation rate.

BAGPIPES accurately recovers $A_V$ and $\delta$ distributions.

The observed $A_V$-$\delta$ correlation is not due to fitting artifacts.

Abstract

Fitting model spectral energy distributions (SED) to galaxy photometric data is a widely used method to recover galaxy parameters from galaxy surveys. However, the parameter space used to describe galaxies is wide and interdependent, and distinctions between real and spurious correlations that are found between these parameters can be difficult to discern. In this work, we use the SED fitting code BAGPIPES to investigate degeneracies between galaxy parameters and the effect of the choice of different sets of photometric bands. In particular, we focus on optical to infrared wavelength coverage, and on two parameters describing the galaxies' dust attenuation law: $A_V$ and $\delta$, which characterize dust column density and the slope of a flexible dust attenuation law, respectively. We demonstrate that 1) a degeneracy between the residual (the difference between truth and recovered value) $A_V$ and star formation rate exists, but this is lifted when WISE bands are included; 2) BAGPIPES is able to accurately recover the input $A_V$ and $\delta$ distributions and relations (differences in slope of less than 1.7$\sigma$ for a flat relation, less than 1.2$\sigma$ for an observationally-motivated relation from Salim et al. 2018) and is not introducing spurious correlations between these parameters. Our findings suggest that the information needed to constrain $A_V$ and $\delta$ well enough individually exists in the data, especially when IR is added. This indicates that recent works finding a correlation between $A_V$ and $\delta$ are not being misled by fitting degeneracies from their SED fitting code.