The THESAN-ZOOM project: Burst, quench, repeat -- unveiling the evolution of high-redshift galaxies along the star-forming main sequence

TL;DR

This work analyzes the high-redshift SFMS and its scatter using the THESAN-ZOOM simulations, revealing an intrinsic normalization evolution scaling as $\propto (1+z)^{\mu}$ with $\mu = 2.64 \pm 0.03$ for SFR on 10 Myr timescales and a weak stellar-mass dependence. By forward-modeling H$\alpha$ and UV tracers, the authors show that LyC escape and bursty SFHs bias observational inferences, necessitating revised SFR calibrations (SFR$_8$ and SFR$_{24}$) and highlighting tracer-dependent timescales ($\sim$7–9 Myr for H$\alpha$, $\sim$22–31 Myr for UV). They identify two burst modes—externally driven inflows from the CGM and mergers, and internally driven breathing cycles within galaxies—explaining why short-term burstiness increases at high redshift while total SFMS scatter grows more slowly due to longer-term environmental effects. The results emphasize CGM inflow variability as a primary driver of population-wide SFMS scatter and burstiness, with observational biases and sample completeness playing crucial roles in interpreting high-z galaxy growth. Overall, THESAN-ZOOM provides predictions for intrinsic SFMS evolution and burst-driven variability that inform JWST-era studies and motivate larger-volume simulations to span cosmic noon and beyond.

Abstract

Characterizing the evolution of the star-forming main sequence (SFMS) at high redshift is crucial to contextualize the observed extreme properties of galaxies in the early Universe. We present an analysis of the SFMS and its scatter in the THESAN-ZOOM simulations, where we find a redshift evolution of the SFMS normalization scaling as $\propto (1+z)^{2.64\pm0.03}$, significantly stronger than is typically inferred from observations. We can reproduce the flatter observed evolution by filtering out weakly star-forming galaxies, implying that current observational fits are biased due to a missing population of lulling galaxies or overestimated star-formation rates. We also explore star-formation variability using the scatter of galaxies around the SFMS ($σ_{\mathrm{MS}}$). At the population level, the scatter around the SFMS increases with cosmic time, driven by the increased importance of long-term environmental effects in regulating star formation at later times. To study short-term star-formation variability, or ''burstiness'', we isolate the scatter on timescales shorter than 50 Myr. The short-term scatter is larger at higher redshift, indicating that star formation is indeed more bursty in the early Universe. We identify two starburst modes: (i) externally driven, where rapid large-scale inflows trigger and fuel prolonged, extreme star formation episodes, and (ii) internally driven, where cyclical ejection and re-accretion of the interstellar medium in low-mass galaxies drive bursts, even under relatively steady large-scale inflow. Both modes occur at all redshifts, but the increased burstiness of galaxies at higher redshift is due to the increasing prevalence of the more extreme external mode of star formation.

Figures (20)

• Figure 1: Three massive thesan-zoom galaxies, m12.2 subhalo 0 (the two left columns), m11.5 subhalo 0 (the two central columns), and m11.1 subhalo 0 (the two right columns) in H$\alpha$ ($1^\text{st}$, $3^\text{rd}$, and $5^\text{th}$ column) and UV ($2^\text{th}$, $4^\text{th}$, and $6^\text{th}$ column) emission from $z=3$ to $z=8$ top to bottom. All images are 120 ckpc by 120 ckpc, and a bar in the lower left of each row shows the scale in pkpc. Luminosity is scaled between zero and the 99.9th percentile brightest pixel in an image. The galaxies show remarkable variation across cosmic time, including distortions due to galaxy--galaxy interactions. Offsets between the UV and H$\alpha$ emission can arise due to a variety of effects, including radiative transfer processes and sites of star formation being spatially disconnected from older stars McClymont:2025ab.
• Figure 2: The star-forming main sequence (SFMS) for the thesan-zoom galaxies across six redshift bins. The shading shows the log-scale number density of galaxies. The solid blue line represents our best fit to the SFMS using the intrinsic values for SFR$_{100}$, whereas the cyan dashed lines show our best fit to the SFR$_\mathrm{UV}$ derived from the intrinsic UV emission (1500Å). The upper dashed black line shows the one-to-one line, corresponding to all stars having formed within the averaging time, whereas the lower dashed black line corresponds to the value of $m_{\mathrm{sSFC}}$ used in the fitting. Our best fit to the intrinsic SFMS$_{100}$ shows a lower redshift dependence than for SFMS$_{10}$, which is likely due to the bunching up of galaxies on the one-to-one line at high redshift for the 100 Myr. An alternative way to view this effect is that most galaxies have rising SFHs at high redshift, whereas there is an equilibrium of rising and falling SFHs at lower redshift. Additionally, a higher normalization of the SFMS at longer averaging timescales is expected for bursty star formation Caplar:2019aa, so this trend is also consistent with increasingly bursty star formation with redshift. This requires that the shorter-timescale SFRs must evolve more rapidly with redshift. The apparent agreement between SFMS$_\mathrm{UV}$ and SFMS$_\mathrm{100}$ is actually serendipitous. The UV best traces SFR$_{24}$ which evolves more rapidly with redshift, but is also biased to a flatter redshift evolution by various effects (see text for details).
• Figure 3: The same as Fig. \ref{['fig:sfms']} but for SFR$_{10}$ and SFR$_{\mathrm{H}\alpha}$ derived from the intrinsic H$\alpha$ emission. Our SFMS$_{10}$ fit shows remarkably good agreement with expectations from dark matter halo growth in DMO simulations. This implies that overall galaxy evolution is largely regulated and described by hierarchical structure growth despite the myriad baryonic processes included in our simulations.
• Figure 4: The relationship between intrinsic and observed SFRs. The observed SFRs use traditional SFR calibrations for H$\alpha$ (top panel) and UV (1500Å, bottom panel). The black line shows the one-to-one line, where all galaxies would lie if the calibration was perfect. The black points show the median values calculated in bins, with the y-axis errors showing the $16^\text{th}$--$84^\text{th}$ range, and the x-axis errors showing the bin range. SFR$_{\mathrm{H}\alpha}$ is systematically lower than SFR$_{10}$ because effects such as LyC escape and dust absorption of LyC tend to decrease H$\alpha$ flux as well as causing significant scatter. SFR$_{\mathrm{UV}}$ is systematically higher than SFR$_{100}$ and is highly scattered, primarily due to the star-formation histories varying on shorter timescales than 100 Myr.
• Figure 5: The evolution of SFR offsets from the main sequence ($\Delta\mathrm{MS}$) for an individual galaxy (m12.2, subhalo 0). The offsets from the intrinsic main sequences behave as expected: $\Delta\mathrm{MS}_{100}$ acts as a damped and lagging companion to $\Delta\mathrm{MS}_{10}$. Galaxies with high $\Delta\mathrm{MS}_{10}$ and low $\Delta\mathrm{MS}_{100}$ are beginning a burst, and those with high $\Delta\mathrm{MS}_{100}$ and low $\Delta\mathrm{MS}_{10}$ are recently quenched. The observed picture is more complex. $\Delta\mathrm{MS}_{\mathrm{H}\alpha}$ traces $\Delta\mathrm{MS}_{10}$ well because systemic issues in the SFR$_{\mathrm{H}\alpha}$ to SFR$_{10}$ conversion are less important for SFMS offsets compared to absolute SFR values. However, $\Delta\mathrm{MS}_{\mathrm{UV}}$ (1500Å) traces $\Delta\mathrm{MS}_{100}$ poorly. This is primarily due to the short nature of the bursts, which leads to a rapidly changing UV to SFR$_{100}$ conversion factor as generations of stars age.