Variability of the UV luminosity function with SPICE

We investigate the variability of the UV luminosity function (UVLF) at $z > 5$ using the SPICE suite of cosmological, radiation-hydrodynamic simulations, which include three distinct supernova (SN) feedback models: bursty-sn, smooth-sn, and hyper-sn. The bursty-sn model, driven by intense and episodic SN explosions, produces the highest fluctuations in the star formation rate (SFR). Conversely, the smooth-sn model, characterized by gentler SN feedback, results in minimal SFR variability. The hyper-sn model, featuring a more realistic prescription that incorporates hypernova (HN) explosions, exhibits intermediate variability, closely aligning with the smooth-sn trend at lower redshifts. These fluctuations in SFR significantly affect the $\rm{M_{UV} - M_{halo}}$ relation, a proxy for UVLF variability. Among the models, bursty-sn produces the highest UVLF variability, with a maximum value of 2.5. In contrast, the smooth-sn and hyper-sn models show substantially lower variability, with maximum values of 1.3 and 1.5, respectively. However, in all cases, UVLF variability strongly correlates with host halo mass, with lower-mass halos showing greater variability due to more effective SN feedback in their shallower gravitational wells. The bursty-sn model, though, results in higher amplitudes. Variability decreases in lower mass haloes with decreasing redshift for all feedback models. This study underscores the critical role of SN feedback in shaping the UVLF, and highlights the mass and redshift dependence of its variability, suggesting that UVLF variability may alleviate the bright galaxy tension observed by JWST at high redshifts.

💡 Research Summary

This paper investigates how supernova (SN) feedback shapes the variability of the ultraviolet luminosity function (UVLF) of galaxies at redshifts greater than five, using the SPICE suite of radiation‑hydrodynamic cosmological simulations. Three distinct SN feedback prescriptions are examined: “bursty‑sn,” where each stellar particle experiences a single, energetic SN event 10 Myr after formation; “smooth‑sn,” which distributes SN explosions over a 3–40 Myr interval according to progenitor mass; and “hyper‑sn,” which adopts the same timing as smooth‑sn but draws the injected energy from a normal distribution (mean 1.2 × 10⁵¹ erg, range 10⁵⁰–2 × 10⁵¹ erg) and adds a metallicity‑dependent hypernova component (10⁵² erg). All other simulation parameters—box size (10 h⁻¹ cMpc), dark‑matter particle mass (6.38 × 10⁵ M⊙), adaptive‑mesh refinement down to ~28 pc at z = 5, star‑formation sub‑grid turbulence model, and multi‑group radiation transport—are held constant, allowing a clean comparison of feedback effects.

The authors compute intrinsic 1500 Å UV luminosities from BPASSv2.2.1 stellar population synthesis models (Chabrier IMF) and apply dust attenuation using the Monte‑Carlo radiative‑transfer code RASCAS, which accounts for metallicity‑ and ionisation‑dependent dust densities. Intrinsic and dust‑attenuated UVLFs are presented at z ≈ 11–12 and at lower redshifts (z ≈ 10, 8, 6) in time‑slices of 10–50 Myr to capture short‑term variability.

Key findings:

1. SFR variability – The bursty‑sn model produces the largest star‑formation rate (SFR) fluctuations because the sudden injection of 2 × 10⁵¹ erg per SN drives strong, episodic gas outflows that temporarily quench star formation before gas re‑accretes. Smooth‑sn yields the most stable SFR, while hyper‑sn shows intermediate stochasticity due to its broader energy distribution and occasional hypernovae.
2. UVLF variability – The amplitude of UVLF fluctuations, quantified as the ratio of the maximum to minimum φ in a given magnitude bin, reaches ≈2.5 for bursty‑sn, ≈1.5 for hyper‑sn, and ≈1.3 for smooth‑sn. The variability is strongest at the faint end (M_UV ≈ −15) where dust attenuation first becomes significant, and it declines toward brighter magnitudes where the galaxy population is dominated by a few massive systems.
3. Mass dependence – Variability correlates strongly with host halo mass. Low‑mass halos (M_halo ≈ 10⁹–10¹⁰ M⊙) exhibit the largest UVLF scatter because their shallow potential wells allow SN feedback to expel gas efficiently, leading to bursty star formation. In massive halos (M_halo > 10¹¹ M⊙) the deeper potential well damps feedback effects, reducing both SFR and UVLF variability.
4. Redshift evolution – Across all models, the magnitude of UVLF variability declines with decreasing redshift. At higher redshifts (z ≈ 12–10) population growth and feedback‑driven stochasticity dominate, while by z ≈ 6 the galaxy population has largely settled into a smoother growth regime.
5. Comparison with observations – When dust attenuation is included, the simulated UVLFs match current HST and JWST measurements over the magnitude range probed by the 10 cMpc box (M_UV ≈ −12 to −20), except that the hyper‑sn model underpredicts the faintest counts. The bursty‑sn model reproduces the observed bright‑end excess better, suggesting that strong, episodic feedback could alleviate the “bright galaxy tension” reported by JWST at z > 10.

The authors also examine the stellar‑mass distribution of galaxies contributing to the bright end (M_UV ≤ −17). Even galaxies with stellar masses as low as 10⁶·⁵–10⁷ M⊙ can appear in the bright UV bin, especially in bursty‑sn runs where temporary SFR spikes boost UV output. Median stellar masses of bright galaxies shift from ≈10⁸ M⊙ at z ≈ 10 to ≈10⁹ M⊙ at z ≈ 6, consistent with semi‑analytic expectations (Mason et al. 2015, 2023).

In summary, the study demonstrates that the nature of SN feedback—its timing, energy budget, and inclusion of hypernovae—directly governs the stochasticity of star formation and, consequently, the variability of the high‑z UVLF. Strong, bursty feedback can produce UVLF fluctuations large enough to reconcile theoretical predictions with the unexpectedly high number densities of bright galaxies observed by JWST, while smoother feedback leads to more modest variability. The work highlights the importance of incorporating realistic, stochastic SN feedback in cosmological simulations to interpret early‑universe galaxy surveys. Future extensions should explore larger simulation volumes, alternative initial mass functions, and varying star‑formation efficiencies to further test the robustness of these conclusions.