The evolution of the dust grain size distribution has been studied in recent years with great detail in cosmological hydrodynamical simulations taking into account all the channels under which dust evolves in the interstellar medium. We present a systematic analysis of the observed spectral energy distribution of a large sample of galaxies in the local Universe in order to derive not only the total dust masses but also the relative mass fraction between small and large dust grains (DS/DL). Simulations reproduce fairly well the observations except for the high-stellar mass regime where dust masses tend to be overestimated. We find that ∼45 per cent of galaxies exhibit DS/DL consistent with the expectations of simulations, while there is a subsample of massive galaxies presenting high DS/DL (log (DS/DL) ∼ −0.5), and deviating from the prediction in simulations. For these galaxies which also have high-molecular gas mass fractions and metallicities, coagulation is not an important mechanism affecting the dust evolution. Including diffusion, transporting large grains from dense regions to a more diffuse medium where they can be easily shattered, would explain the observed high DS/DL values in these galaxies. With this study, we reinforce the use of the small-to-large grain mass ratio to study the relative importance of the different mechanisms in the dust life cycle. Multiphase hydrodynamical simulations with detailed feedback prescriptions and more realistic subgrid models for the dense phase could help to reproduce the evolution of the dust grain size distribution traced by observations.

Attenuation by dust severely impacts our ability to obtain unbiased observations of galaxies, especially as the amount and wavelength dependence of the attenuation varies with the stellar mass M*, inclination i, and other galaxy properties. In this study, we used the attenuation – inclination models in ultraviolet, optical, and near-infrared bands designed by Tuffs and collaborators to investigate the average global dust properties in galaxies as a function of M*, the stellar mass surface density μ*, the star-formation rate SFR, the specific star-formation rate sSFR, the star-formation main-sequence offset dMS, and the star-formation rate surface density ΣSFR at redshifts z ∼ 0 and z ∼ 0.7. We used star-forming galaxies from the Sloan Digital Sky Survey (∼20 000) and Galaxy And Mass Assembly (∼2000) to form our low-z sample at 0.04 < z < 0.1 and star-forming galaxies from Cosmological Evolution Survey (∼2000) for the sample at 0.6 < z < 0.8. We found that galaxies at z ∼ 0.7 have a higher optical depth τBf and clumpiness F than galaxies at z ∼ 0. The increase in F hints that the stars of z ∼ 0.7 galaxies are less likely to escape their birth cloud, which might indicate that the birth clouds are larger. We also found that τBf increases with M* and μ*, independent of the sample and therefore redshift. We found no clear trends in τBf or F with the SFR, which could imply that the dust mass distribution is independent of the SFR. In turn, this would imply that the balance of dust formation and destruction is independent of the SFR. Based on an analysis of the inclination dependence of the Balmer decrement, we found that reproducing the Balmer line emission requires not only a completely optically thick dust component associated with star-forming regions, as in the standard model, but an extra component of an optically thin dust within the birth clouds. This new component implies the existence of dust inside H II regions that attenuates the Balmer emission before it escapes through gaps in the birth cloud and we found it is more important in high-mass galaxies. These results will inform our understanding of dust formation and dust geometry in star-forming galaxies across redshift.