The Diversity and Evolution of Dust Attenuation Curves from Redshift z ~ 1 to 9

Abstract

The UV-optical dust attenuation curve is key to interpreting the intrinsic properties of galaxies and provides insights into the nature of dust grains and their geometry relative to stars. In this work, we constrain the UV-optical slope of the stellar attenuation curve using a spectroscopic-redshift sample of ~3800 galaxies at z~1-9, to characterize the diversity and redshift evolution of stellar attenuation curves and to gain insight into dust production and evolution at high redshifts. The sample is constructed from three JWST/NIRCam grism surveys in GOODS and A2744 fields, with a wealth of JWST/NIRCam and HST photometry. With constraints from spectroscopic redshifts and emission line fluxes, we use the Prospector SED fitting code with a flexible dust model. We find that the attenuation curve slope varies strongly with Av at all redshifts, becoming flatter at higher attenuation. We find no strong correlation between attenuation curve slope and size or axis ratio, and the trends with stellar mass and star-formation rate are largely driven by their correlation with Av. We find strong evidence that at fixed Av, the curve becomes flatter with increasing redshift. On average, the attenuation curves derived here are shallower than those at z~0 and than the SMC curve. The highest redshift galaxies at z=7-9 (124 galaxies, a significantly larger sample than in previous studies) show slopes even flatter than the Calzetti curve, implying reduced UV obscuration and lower IR luminosities than expected from an SMC dust curve, by as large as an order of magnitude. Hydrodynamical simulations that couple dust growth to gas chemical enrichment successfully reproduce the different loci of high- and low-redshift galaxies in the slope-Av diagram, suggesting that dust in high-redshift galaxies is increasingly dominated by large grains produced in supernova ejecta with limited ISM processing at early times.

Figures (12)

• Figure 1: Redshift distribution of the sample. Left: Redshift distribution of the parent sample is shown with grey filled histogram. Those with good SED fits ("Good-SED" with reduced $\chi^2 < 2$; 94% of the parent sample) used here are shown with dashed line. Out of the good-SED sample, those with high-significance attenuation curve slope determination (slope value $<1\sigma$ error, where error is marginalized over all other parameters in the SED modeling) are shown with dotted line ("robust-slope" sample). The statistical trends throughout the paper are not significantly changed for the parent sample and the good-SED sample, but for accuracy we show the results with the good-SED sample throughout the paper. Right: The sample is further broken into the survey of origin, for all (solid lines) and those with robust slopes and good fits (dashed lines). The redshift peaks correspond to ranges where specific emission lines are shifted into the observed bands (see §\ref{['sec:data']}). The sharp peaks in the histogram indicate known overdensities in the fields helton24covelo-paz25herad25.
• Figure 2: Comparison of different definitions of attenuation curve slope in the absolute and selective curve formalization, described in §\ref{['sec:slope-def']}, and the Prospector power-law index of old stars dust attenuation curve, "dust_index" (see text). In the middle panel, we show the relation between $\delta$ in Equation \ref{['eq:kappa']} and the absolute curve slope (discussed later in Equation \ref{['eq:slope-conv']}) with a black line. The black curve does not exactly follow the relationship between absolute curve slope and Prospector "dust_index", as the latter is the curve slope only for the old stellar population in the charlotfall00 parametrization.
• Figure 3: Absolute attenuation curve slope ($A_{1500\,\AA}$/$A_V$) as a function of galaxy parameters for objects with high-significance slopes (slope value $<1\sigma$ error), from left to right: optical attenuation ($A_V$), stellar mass, SFR averaged over 50 Myr, and mass-weighted age. Contours represent the average trends of the sample. Individual data points are color-coded by their redshifts. Top histograms show the normalized distribution of the x-axis parameter for the full sample (grey) and in three redshift bins (colored lines). We show the uncertainties of individual points in the left panel, and for simplicity only the median uncertainty of the sample in the rest of the panels (crosses in the top-right corner). These error bars reflect the 16–84th percentiles for the posteriors after marginalizing over all other parameters in the SED modeling. The Spearman correlation factors are shown in each panel (all are significant with p-values $<<1$). The strongest relationship is an anti-correlation with $A_V$. A first-order polynomial in log(slope)–log($A_V$) is shown with a blue curve ($\log(\rm{slope}) = -0.23 \times \log(A_V) + 0.44$). Grey shaded region marks the range of slopes between those of the Calzetti curve (lower bound) and the SMC curve (upper bound).
• Figure 4: Left: Absolute curve slope versus 5500 $\AA$ attenuation and for our data with high-significance slope (black contours) and simulations (blue and red tracks). The black curve is a best-fit polynomial to the data (Figure \ref{['fig:slope_av_params']}). Red and blue points indicate matsumoto25 models with and without dust grain size evolution, respectively. Each track shows the time evolution from 50 Myr to 5 Gyr (in order of increasing size, the points correspond to 50, 100, 250, 500 My, 1, 2, 3, 4, 5 Gyr). The shading of the tracks correspond to inclination angles of 20, 40, 60, 80, and 90$^{\circ}$, such that the darker tracks have lower inclination angles. Middle and right: Optical attenuation $A_V$ and attenuation curve slope as a function of ellipticity (derived from axis ratio, $1-(b/a)$) for observations (FRESCO and CONGRESS datasets only; contours and small dots) and simulations (same symbols as the left panel). Assuming thin disks, the ellipticity parameter traces inclination. We divide the observed data to those at cosmic noon (blue) and higher redshifts (red). While the simulations show dependence of slope and $A_V$ on inclination, there is no significant correlation between the slope or $A_V$ and the inclination in the data. Average uncertainties of the measurements are shown with grey crosses in the panels.
• Figure 5: Panels showing variation in the curve slope as a function of effective radius ($R_e$, top row) and SFR surface density ($\Sigma_{\rm SFR}$, bottom row) in bins of $A_V$ (columns). Blue and red points and contours show the galaxies below and above $z=3$, respectively (limited to the data in GOODS fields, see text). These panels show that, as expected, not only higher redshift galaxies are smaller with higher $\Sigma_{\rm SFR}$ than lower redshift galaxies, but also, there is a correlation between $A_V$ and SFR surface density. However, once $A_V$ is fixed, the curve slope does not change with respect to either size or SFR surface density -- within each redshift range and across the two redshift bins.