NISER-IUCAA New Simulations of JWST GAlaxies and Quasars(NINJA): Properties of galaxies at $5 \leq z \leq 10$

The NINJA suite of cosmological hydrodynamical simulations demonstrates that while specific spectral and dust models can reproduce observed UV luminosity functions for galaxies at $5 \leq z \leq 10$, significant degeneracies between feedback and dust properties, along with resolution limitations at $z > 10$, necessitate higher-resolution simulations and multi-wavelength observations to robustly constrain high-redshift galaxy evolution.

The Gist

Imagine the early universe as a vast, dark construction site just a few hundred million years after the Big Bang. The James Webb Space Telescope (JWST) is like a powerful new crane and camera crew that has finally arrived to take high-definition photos of the very first buildings (galaxies) being constructed.

This paper, titled NINJA, is a report from a team of astronomers who built their own "virtual construction site" using supercomputers. Their goal was to see if their digital models could match the real photos JWST is taking of these ancient galaxies.

Here is a breakdown of what they did and found, using simple analogies:

1. The Virtual Construction Site (The Simulations)

The researchers created three different-sized virtual universes (like building a model city in a shoebox, a living room, and a stadium). They filled these boxes with dark matter and gas, letting gravity pull them together to form galaxies.

• The Challenge: They needed to make sure their digital galaxies looked like the real ones. Specifically, they needed to match how bright these galaxies appear in ultraviolet light (UV), which is the primary way we see young, hot stars.

2. The "Dust Filter" Problem

In the real world, if you try to take a photo of a lightbulb through a dirty window, the light looks dimmer and redder. In space, this "dirty window" is cosmic dust.

• The Issue: The team found that their digital galaxies were naturally too bright and too blue compared to what JWST sees. To fix this, they had to add a "dust filter" to their models.
• The Experiment: They tried different types of "dust recipes." Some recipes assumed dust was made in a simple, straight-line relationship with metal (like mixing paint). Others tried more complex recipes where dust formation changes drastically depending on how "metal-rich" the galaxy is. They also tried different "lenses" (attenuation curves) to see how the dust blocked the light.

3. The "Dust-to-Metal" Ratio (The Secret Ingredient)

To make their virtual galaxies match the real ones, the team had to adjust a dial called $\epsilon$ (epsilon). Think of this as the "dust efficiency knob."

• What they found: They discovered that in the early universe, galaxies were much less efficient at making dust than our own Milky Way is today.
  • At a redshift of 5 or 6 (very early), the dust-to-metal ratio was only about 35% of what we see in our local neighborhood.
  • By redshift 9 or 10 (even earlier), it dropped to less than 10%.
• The Catch: The exact number they needed to turn the knob to depended heavily on which dust recipe they chose. If they changed the recipe, the knob setting changed by a factor of 7! This means we can't be 100% sure exactly how much dust exists yet without more data.

4. The "Baby Stars" Effect (Nebular Emission)

The team realized they were missing a crucial ingredient: Nebular emission.

• The Analogy: Imagine a construction site where the workers (stars) are surrounded by a glowing fog (gas clouds). If you only count the light from the workers, you miss the glow of the fog.
• The Result: When they added the light from this "fog" to their models, the galaxies got brighter, especially the smaller, fainter ones. This helped their models match the real observations much better.

5. The "Top-Heavy" Star Problem (The IMF)

The team also tested what happens if the early universe made "bigger" stars on average than it does today.

• The Analogy: Usually, a star factory makes a mix of small, medium, and large stars (like a standard bakery). But what if the early universe only baked giant loaves of bread?
• The Result: If they assumed the early universe made more massive stars (a "top-heavy" IMF), the galaxies became incredibly bright. This helped explain the faintest galaxies better, but it required even more dust to dim them down to match what JWST sees.

6. The "Too Bright" Problem at the Edge of Time

When they looked at the very earliest galaxies (redshift $z > 10$), their models hit a wall.

• The Issue: Even with their best dust recipes and star assumptions, their virtual galaxies were still too dim compared to the real ones JWST found.
• The Conclusion: The paper suggests their computer models aren't detailed enough yet. It's like trying to draw a high-resolution portrait with a low-resolution pencil; they need a "higher resolution" simulation to understand these earliest galaxies properly.

7. The "Balmer Ratio" and "Color Excess" (The Dust Detective)

The team used specific chemical signatures (like the ratio of two specific colors of light, H-alpha and H-beta) to act as a "dust detective."

• The Finding: They found that the dust around newborn stars (in their "birth clouds") is much redder than the dust floating around the rest of the galaxy.
• The Discrepancy: Their models predicted that the dust around stars and the dust in the rest of the galaxy should be somewhat similar. However, real observations suggest the dust around stars is much more effective at blocking light. This suggests their current "dust recipes" might need a major overhaul.

Summary: What Does This Mean?

The NINJA team successfully built a virtual universe that can mimic the brightness of early galaxies, but only if they carefully tune the amount of cosmic dust and the types of stars being born.

• Dust is key: Even in the very early universe, dust was already forming and dimming the light, but it was much less efficient than it is today.
• We need more data: Because different "dust recipes" give different answers, we need more observations (especially from the ALMA telescope, which looks at dust directly) to figure out the correct recipe.
• We need better computers: To understand the very first galaxies (beyond redshift 10), their current simulations aren't detailed enough. They need to run the simulation with higher resolution to stop the "pixelation" in their models.

In short, the universe was a dusty, star-forming construction site much earlier than we thought, but we are still figuring out exactly how much dust was on the windows and how bright the lights really were.

Technical Summary

Technical Summary: NINJA Simulations of High-Redshift Galaxies (5 ≤ z ≤ 10)

Problem Statement
The James Webb Space Telescope (JWST) has provided unprecedented observations of high-redshift galaxies ($z \gtrsim 5$), revealing details about the epoch of reionization and early galaxy formation. However, discrepancies exist between these observations and predictions from standard $\Lambda$CDM cosmological hydrodynamical simulations. Specifically, observations often indicate an excess of bright galaxies at $z > 10$ and suggest more evolved systems (higher stellar masses and dust content) than simulations tuned to low-redshift data predict. A critical challenge is disentangling the effects of dust attenuation, the Initial Mass Function (IMF), and feedback mechanisms, as these parameters are often degenerate. Current simulations struggle to simultaneously reproduce the Ultraviolet Luminosity Function (UVLF), spectral slopes ($\beta_{UV}$), Balmer line ratios, and dust mass constraints across multiple redshifts without fine-tuning.

Methodology
The authors present the NINJA (NISER-IUCAA New Simulations of JWST GAlaxies and Quasars) suite of cosmological hydrodynamical simulations.

• Simulation Setup: The simulations utilize the MP-GADGET code, employing a pressure-entropy smoothed particle hydrodynamics (SPH) formulation. Three simulation boxes were run with comoving side lengths of 50, 150, and 250 $h^{-1}$ Mpc (L50N2040, L150N2040, L250N2040), each containing $2040^3$ dark matter and gas particles. This configuration allows for the study of a wide range of halo masses while quantifying convergence issues.
• Physics Models: The simulations follow the ASTRID suite's subgrid prescriptions for star formation, supernova feedback, and black hole seeding/growth. Star formation follows a multiphase ISM model, and black holes are seeded in halos above $5 \times 10^9 M_\odot h^{-1}$ with subsequent Bondi-Hoyle-Lyttleton accretion and dual-mode (thermal/kinetic) feedback.
• Post-Processing & Spectral Synthesis:
  • SED Generation: Intrinsic spectra are generated by summing contributions from stellar populations (using bpass-v2.2.1 SSP models) and nebular emission (using CLOUDY).
  • Dust Modeling: Dust attenuation is applied as a post-processing step. The visual extinction ($A_V$) is derived from hydrogen column density ($N_H$) scaled by a dust-to-metal ratio parameter ($\epsilon$) and a metallicity-dependent function $f(Z_g)$.
  • Variations: The study explores multiple prescriptions:
    1. Dust-Metallicity Scaling: Linear vs. broken power-law (double power-law) relations.
    2. Attenuation Curves: Calzetti et al. (2000) vs. SMC-average extinction laws.
    3. Dust Geometry: Single-phase (ISM only) vs. two-phase (ISM + "birth cloud" attenuation for stars < 10 Myr).
    4. IMF: Standard Chabrier IMF vs. a top-heavy Kroupa IMF (KP300-TH) with an upper mass cutoff of $300 M_\odot$.
• Calibration: For each redshift ($5 \le z \le 10$), the parameter $\epsilon$ is tuned via $\chi^2$ minimization to match the observed UV Luminosity Function (UVLF).

Key Results

1. Reproduction of UVLF: The fiducial models, when calibrated with appropriate $\epsilon$ values, successfully reproduce the observed UVLFs over $5 \le z \le 10$. The inclusion of nebular continuum emission systematically boosts the UV luminosity, improving agreement with observations, particularly at the faint end.
2. Dust-to-Metal Ratio Evolution: The inferred dust-to-metal ratio decreases with redshift. For the fiducial model (Calzetti curve), the ratio at $z=5-6$ is $\sim35\%$ of the local Milky Way value, dropping to $\le 10\%$ at $z \ge 9$. However, this normalization varies by a factor of $\sim7$ depending on the adopted attenuation curve and dust-metallicity scaling.
3. Degeneracies and Constraints:
   • Attenuation Curves: While both Calzetti and SMC curves can fit the UVLF, they predict different B-band luminosity functions, H$\alpha$ luminosities, and $\beta_{UV}$ slopes. The observed $\beta_{UV}$ vs. $M_{UV}$ relation at $z=5-7$ favors steeper extinction curves (closer to SMC) over the flatter Calzetti law.
   • IMF Effects: A top-heavy IMF (KP300-TH) produces more UV light per unit star formation rate, requiring higher dust-to-metal ratios to match the UVLF. This IMF helps fit the faint-end UVLF at $7 \le z \le 9$ but implies dust properties closer to local values at early epochs.
   • Balmer Ratios: Models including "birth cloud" attenuation produce higher nebular color excess ($E(B-V)_{neb}$) relative to stellar color excess ($E(B-V)_{star}$), consistent with some high-redshift trends, though the ratio $f = E(B-V)_{star}/E(B-V)_{neb}$ remains difficult to constrain precisely with current data.
4. High-Redshift Limits ($z > 10$): The simulations underpredict the UVLF at $z \ge 10$ relative to current observations, even when adopting a top-heavy IMF and neglecting dust. The authors attribute this to a lack of numerical convergence; the number of galaxies in the simulation boxes drops significantly at $z \sim 15$, indicating that higher-resolution simulations are required to robustly model galaxies at these redshifts.
5. H$\alpha$ Luminosity: The simulations systematically overpredict the H$\alpha$ luminosity function compared to observations at $z \sim 5-6$. The authors suggest this discrepancy could be resolved by including dust within the HII regions (currently assumed dust-free in the sub-grid model) or by allowing a higher escape fraction of ionizing photons.

Significance and Claims
The paper claims that the NINJA suite provides a comprehensive framework for interpreting JWST data by systematically varying dust, IMF, and feedback parameters. The primary significance lies in demonstrating that:

• Dust is crucial even at high $z$: Non-negligible dust attenuation is required to match the shape of the UVLF at the bright end, implying efficient dust formation in early galaxies.
• Multi-observable constraints are necessary: Simultaneously reproducing the UVLF, B-band LF, H$\alpha$ LF, and spectral slopes is essential to break degeneracies between dust models and IMF assumptions.
• Resolution limits: Current discrepancies at $z > 10$ highlight the need for higher-resolution simulations to avoid numerical convergence issues.
• Future Directions: The authors emphasize that independent constraints from ALMA (dust mass) and JWST spectroscopy (Balmer ratios, Ly$\alpha$ damping wings) are critical for refining dust models. They note that their current results are preliminary regarding the faint-end UVLF and H$\alpha$ luminosity, which will be addressed in future work involving varied feedback parameters.

The paper concludes that while the models can be tuned to fit specific observables (like the UVLF), the resulting physical parameters (dust-to-metal ratios, attenuation curves) vary significantly based on the chosen prescriptions, underscoring the need for multi-wavelength, multi-observable data to constrain galaxy formation physics in the early universe.