A physically-motivated template set for high-z galaxy SED fitting

TL;DR

This work addresses the challenge of extracting physical properties from high-redshift galaxy SEDs by constructing a physically-motivated template set derived from the galaxy-formation code ares. The templates are made redshift-dependent and augmented with a burst component, and are integrated into the SED-fitting tool EAZY to yield not only redshifts but full star-formation histories, SFRs, and stellar masses. Validation against the Santa Cruz semi-analytic model shows accurate redshift recovery and closer agreement on stellar masses than standard EAZY templates, while application to JWST/JADES data reveals that $z>8$ galaxies are often bursty, form stars earlier than some models predict, and follow a rising star-forming main sequence with redshift. The framework enables rapid, physically grounded inference and direct comparisons to theoretical models, guiding improvements in galaxy formation scenarios for the cosmic dawn.

Abstract

We introduce a new physically-motivated spectral template set for fitting the spectral energy distributions (SEDs) of high-z galaxies. We use the public galaxy formation code ARES to generate star formation histories of thirteen representative galaxies with diverse masses and generate their predicted spectra across a set of redshifts at z > 6. The model parameters are calibrated to reproduce the properties of z > 6 galaxies observed by HST. Motivated by the apparent importance of bursty star formation at high redshifts, we also include templates with recent starbursts. We use these templates with the SED-fitting code EAZY to analyze both an independent theoretical model and a public sample of JWST-observed galaxies from the JADES survey. The comparison with a semi-analytic model demonstrates that our fitting framework accurately measures the galaxy properties, even when the underlying assumptions of the model differ from ours. Our preliminary application to JWST data shows that galaxies at z > 8 are often bursty (especially at small galaxy masses), follow a star-forming main sequence similar to those at lower redshift (albeit with a higher normalization), and form stars earlier than expected in ares. Our SED-fitting framework is very fast (thanks to the efficiency of EAZY) but provides full inferred star formation histories for each source. Additionally, it enables a direct comparison to theoretical models and helps point toward improvements necessary in those models.

Figures (14)

• Figure 1: The star formation histories (left) and stellar mass growth histories (right) of our template galaxies as a function of redshift. The colors are the same in both panels for each galaxy; at $z=6$, the total halo masses range from $\sim10^{8}$--$10^{12} \; \mathrm{M_{\odot}}$. Creating a complete template set with a variety of star formation histories is necessary to obtain high quality fits, considering that in ares low-mass galaxies grow differently than high-mass galaxies, as do high and low-redshift galaxies. The templates span $z = 6$--$19$.
• Figure 2: Left: The star-forming galaxy main sequence (i.e., star formation rate vs. stellar mass) for the ares galaxies at $z = 6$, $z = 12$, and $z = 15$, in purple, orange, and gray dashed lines, respectively. The dots show the location of each template galaxy at $z=6$ and 12 (the bottommost purple dots correspond to galaxies that have not formed stars before $z = 12$), and the solid lines between the dots indicate the paths the galaxies evolve. Because galaxies at higher redshifts have less time to reach a fixed stellar mass than their lower-$z$ counterparts, they must form stars more quickly to reach that stellar mass, and thus have higher SFRs. Therefore, the star-forming main sequence evolves across redshifts, suggesting that fitting galaxies at one redshift from templates from another will result in errors. Right: Distribution of mean stellar ages, defined as the half-mass time, or how long ago $50\%$ of stars were formed in each galaxy, for the $z = 6$ and $z = 12$ template galaxies.
• Figure 3: Left: The rest spectra of each template at $z = 9$, normalized to the Lyman-$\alpha$ break at $\lambda = 1216 \ \text{\AA}$. The colors match the galaxies in figure \ref{['fig:sfh-mass']}, except the most blue object in figure \ref{['fig:sfh-mass']} is not included here because it has not yet begun forming stars at $z = 9$. The varying spectral shape across halo masses point to the need for a variety of high-$z$ spectra in SED fitting. Note that we have assumed the IGM absorbs all light blueward of Ly$\alpha$ here (as EAZY assumes for high-$z$ galaxies). Right: An example rest spectrum of a galaxy in our template set (red) with the inclusion of a SSP burst 5 Myr old (gold) which makes up $33\%$ of the mass of the template. The burst spectrum is isolated in blue. Note that these spectra have Ly$\alpha$ and the Lyman continuum included.
• Figure 4: True redshifts of SC SAM z > 8 galaxies against the redshift fitting from our template set (orange), from Larson2023 (green), and from Hainline2023 (blue). The agreement between the three template sets is good; all accurately reproduce the input redshifts within $\Delta z \sim 0.1$, although our templates appear to be somewhat less biased. Note that the poor fits at $z \sim 10$ are due to a small gap in JWST filter coverage for the Lyman break at this redshift.
• Figure 5: $\chi^{2}$ cumulative distributions from these fits for each template set, including EAZY's default template set ("tweak_fsps_QSF_12_v3.param") in black. The different template sets qualititatively have the same $\chi^{2}$ distribution (or quality of fits).