The AGORA High-resolution Galaxy Simulations Comparison Project. X: Formation and Evolution of Galaxies at the High-redshift Frontier

1. Problem Statement

Recent observations by the James Webb Space Telescope (JWST) have revealed unexpectedly luminous and massive galaxies at the "Cosmic Dawn" (redshifts $z \ge 10$). These galaxies exhibit stellar masses and UV luminosities significantly higher than predicted by standard $\Lambda$CDM cosmological models. This creates a tension in galaxy formation theories, suggesting either:

• A breakdown of standard physics at high redshifts.
• The need for exotic mechanisms (e.g., top-heavy Initial Mass Functions, feedback-free starbursts, or AGN dominance).
• Or, that standard models, when applied correctly with high resolution, can reproduce these observations without new physics.

The paper aims to resolve this tension by testing whether standard galaxy formation models, calibrated at lower redshifts, can reproduce high-redshift observations when applied across multiple, state-of-the-art numerical simulation codes.

2. Methodology

The study utilizes the AGORA High-z Run, a suite of cosmological "zoom-in" simulations targeting massive dark matter halos ($10^{10} - 10^{11} M_\odot$ at $z=10$).

• Participating Codes: Six distinct hydrodynamical codes were employed to ensure robustness against numerical artifacts:
  • Mesh-based (AMR): ENZO, RAMSES.
  • Particle-based (SPH): CHANGA, GADGET-3, GADGET-4.
  • Hybrid (MFM/MFV): GIZMO.
• Initial Conditions & Resolution:
  • Five target halos were selected from a $64 \, (Mpc/h)^3$ box, refined to an effective resolution of $4096^3$ cells/particles.
  • Physical Resolution: Uniform resolution of 22–32 pc during peak star formation epochs.
  • Refinement Strategy: Mesh codes reached $348$ comoving pc; particle codes used softening lengths of $1600$ comoving pc until $z=19$, then $80$ physical pc.
• Subgrid Physics (AGORA CosmoRun Framework):
  • Chemistry/Cooling: GRACKLE library with CLOUDY tables; metallicity floor of $10^{-4} Z_\odot$.
  • Star Formation: Schmidt law ($\epsilon_\star = 0.01$) triggered at $n_H > 1 \, \text{cm}^{-3}$.
  • Feedback: Each code implemented its own standard stellar feedback (thermal, kinetic, delayed cooling, superbubbles) without tuning specifically for high-$z$.
  • Dust: Modeled via SKIRT radiative transfer code. Dust density scales with gas density and metallicity, with a low dust-to-metal ratio ($f_{dust} = 0.1$) appropriate for high-$z$.
• Mock Observations: SKIRT was used to generate mock JWST NIRCam images and calculate UV luminosities, accounting for dust attenuation and nebular emission.

3. Key Contributions

• First Multi-Code Comparison at $z \ge 10$: This is the first systematic comparison of six major simulation codes specifically targeting the Cosmic Dawn, moving beyond previous comparisons limited to Milky Way-mass halos at $z=0$.
• Validation of "Vanilla" $\Lambda$CDM: It tests whether standard models (calibrated at low-$z$) can explain JWST data without invoking new physics (e.g., top-heavy IMFs or feedback-free starbursts) as a primary driver.
• Quantification of Inter-Code Variance: It isolates how different numerical solvers (AMR vs. SPH vs. Hybrid) and feedback implementations affect high-redshift galaxy properties.

4. Key Results

A. Stellar Mass and Halo Mass

• Convergence: Stellar mass histories show broad convergence across codes for massive halos ($M_{halo} \ge 5 \times 10^{10} M_\odot$), with differences generally within $\sim 1$ dex.
• Observational Match: Massive halos successfully reproduce the stellar masses and UV luminosities of observed JWST galaxies at $10 \le z \le 12$ without additional subgrid physics.
• High-$z$ Discrepancy: At $z > 12$, the simulations tend to underpredict stellar masses and UV luminosities compared to the most extreme JWST candidates, suggesting that either higher resolution or specific high-$z$ physics may be needed for the very earliest epochs.

B. Metallicity

• Divergence: Unlike stellar mass, metallicity shows significant inter-code variation (up to $\sim 1$ dex).
• Feedback Sensitivity: Codes with strong outflows (RAMSES, CHANGA, GADGET-3) mix metals into the circumgalactic medium (CGM), while codes with weaker outflows (ENZO, GIZMO) keep metals concentrated in the center.
• Observational Match: The simulations generally match JWST metallicity data at $z \le 12$, but underpredict it at higher redshifts (except for RAMSES, which matches well at $z \sim 13$).

C. Star Formation Efficiency (SFE) & Feedback-Free Starbursts (FFB)

• Bursty Star Formation: All codes exhibit bursty star formation.
• SFE Variance: Particle-based codes (GIZMO, CHANGA) often achieve higher instantaneous SFEs ($\epsilon_{inst} \sim 1$) in dense cores compared to mesh-based codes.
• FFB Signatures: Some massive halos reach gas densities ($n_H > 3000 \, \text{cm}^{-3}$) where feedback becomes inefficient, mimicking the "Feedback-Free Starburst" scenario. However, the current resolution ($\sim 30$ pc) is insufficient to resolve the internal structure of these clouds to definitively validate the FFB model.

D. Dust Attenuation

• Critical Factor: The presence of dust is the dominant factor suppressing UV luminosity, not the precise dust-to-metal ratio.
• Impact: Removing dust increases UV magnitude by nearly 2 magnitudes at $z=10$.
• Ratio Sensitivity: Varying the dust-to-metal ratio (0.01 to 0.4) produces only mild differences.
• Code Differences: Particle-based codes produce more dust mass due to denser gas cores, leading to stronger attenuation than mesh-based codes.

5. Significance and Limitations

• Significance: The study suggests that the "tension" between JWST observations and $\Lambda$CDM may be less severe than initially thought. Standard galaxy formation models can reproduce observed high-$z$ galaxies if the host halos are sufficiently massive. The results highlight that numerical resolution and feedback implementation are the primary sources of uncertainty, rather than a fundamental failure of the cosmological model.
• Limitations:
  • Resolution: The $\sim 30$ pc resolution cannot resolve Giant Molecular Clouds (Jeans length $< 10$ pc), preventing direct validation of small-scale fragmentation or the FFB mechanism.
  • Missing Physics: The simulations do not include AGN feedback or primordial non-equilibrium chemistry (though the latter is expected to be sub-dominant for these massive halos).
  • Halo Mass Matching: Direct comparison with observations is limited because the halo masses of observed JWST galaxies are unknown.

Conclusion

The AGORA High-z Run demonstrates that massive halos ($M_{halo} \ge 5 \times 10^{10} M_\odot$) formed in standard $\Lambda$CDM cosmology can reproduce the stellar masses, metallicities, and UV luminosities of JWST galaxies at $z \le 12$. While inter-code differences in metallicity and feedback implementation remain significant, the overall convergence supports the viability of standard galaxy formation models at Cosmic Dawn. Future work will focus on higher-resolution simulations ($\le 10$ pc) to probe the internal structure of these early galaxies and the role of AGN.