A Selection Aware View of Black Hole-Galaxy Coevolution at High Redshift

Using a selection-aware Bayesian framework on JWST JADES data, this study reveals that the black hole-stellar mass relation at high redshift ($z \gtrsim 4$) shares the same slope and normalization as the local Universe but exhibits significantly larger intrinsic scatter, suggesting that the mean scaling was established early while the diversity of growth histories remains high.

The Gist

Imagine the early universe as a bustling, chaotic construction site. In the center of every massive building (a galaxy), there is a giant, invisible engine (a supermassive black hole). For decades, astronomers have known that in our quiet, modern neighborhood (the nearby universe), the size of the engine is perfectly matched to the size of the building. Big buildings have big engines; small buildings have small engines. They grow up together in a tight, synchronized dance.

But what about when the universe was a toddler, just a billion years old? Did these engines and buildings grow together back then, or was it a messy, uncoordinated mess?

This paper, written by a team of astronomers using the powerful James Webb Space Telescope (JWST), tries to answer that question. Here is the story of their discovery, explained simply.

The Problem: The "Brightest Kid" Bias

The team looked at a group of ancient galaxies using JWST. They wanted to measure the size of the black holes inside them. However, there was a catch: JWST can only see the loudest, brightest voices.

Imagine trying to guess the average height of everyone in a school by only measuring the students standing on top of the bleachers. You would think everyone is a giant, because you missed all the kids standing on the ground.

In astronomy, this is called a selection bias. The black holes that are easiest to see are the ones that are eating a lot of gas and shining very brightly. The smaller, quieter black holes are hidden in the noise. If you just look at the bright ones, you might think black holes were "overgrown" giants in the early universe, much bigger than they should be for their host galaxies.

The Solution: The "Invisible Net"

Instead of just looking at the data they had, the authors built a virtual simulation.

Think of it like this: They created a giant digital sandbox with millions of fake galaxies and black holes. They knew the "true" rules of how big the black holes should be. Then, they ran their "JWST camera" through this sandbox to see what it would actually detect.

They realized that the camera acts like a net with holes.

• If a black hole is huge and bright, the net catches it easily.
• If a black hole is small or the galaxy is very dusty, the net lets it slip through.

By mapping exactly where the holes in the net were, they could mathematically "fill in the gaps." They essentially asked, "If we could see the ones we missed, what would the whole picture look like?"

The Discovery: Same Average, Wilder Variance

When they corrected for the "holes in the net," they found something surprising.

1. The Average is Normal:
The average relationship between the black hole and the galaxy in the early universe is actually the same as it is today. The "dance" was already synchronized. The black holes weren't systematically overgrown; they were just the right size for their galaxies, on average.

2. The Chaos is Real:
However, the variety was much wilder.

• Today: It's like a well-organized marching band. Everyone is in step. If you know the size of the building, you can predict the engine size with high precision.
• The Early Universe: It was like a jazz jam session. While the average volume was the same, some bands were playing incredibly loud, and others were barely whispering.

The authors found that the "scatter" (the spread of data points) was huge. Some black holes were way too big for their galaxies, and others were way too small.

Why the Chaos?

Why was the early universe so messy? The authors suggest a few reasons, using a cooking analogy:

• Bursty Eating: In the early universe, black holes didn't eat a steady meal. They had "feast or famine" cycles. Sometimes they gorged on gas for a short time (getting fat quickly), and then starved. This made their sizes fluctuate wildly compared to their galaxies.
• Different Recipes: Maybe the "seeds" (the baby black holes) started at different sizes. Some started as tiny pebbles, others as boulders. It takes time for the universe to smooth out these differences.
• No Traffic Control: The feedback loops that usually keep black holes and galaxies in sync (where the black hole blows gas away to stop itself from growing too fast) might have been lagging or delayed back then.

The Takeaway

The paper tells us that the fundamental rule connecting black holes and galaxies was established very early in the universe's history. The "average" was already set.

However, the early universe was a wilder place. The growth of black holes and galaxies was less synchronized, more chaotic, and more diverse than it is today. The "overmassive" black holes we see in early data weren't a sign that the rules had changed; they were just the lucky few that happened to be caught in the act of a "feast," while the quieter ones were missed by our telescopes.

In short: The universe had the right blueprint from the start, but the construction was a bit more chaotic and less organized than the neat, quiet neighborhood we live in today.

Technical Summary

Here is a detailed technical summary of the paper "A Selection Aware View of Black Hole–Galaxy Coevolution at High Redshift" by Ziparo et al. (2026).

1. Problem Statement

The James Webb Space Telescope (JWST) has revealed a large population of broad-line Active Galactic Nuclei (AGN) at redshifts $z \gtrsim 4$, offering a new window into the co-evolution of supermassive black holes (SMBHs) and their host galaxies during the first gigayear of cosmic history. However, previous studies attempting to constrain the black hole mass–stellar mass ($M_{BH}–M_\star$) relation at these epochs have faced significant challenges:

• Selection Biases: Current samples are heavily flux-limited, favoring the most luminous, high-accretion sources. This creates a "Malmquist bias" (or Lauer bias) where low-mass black holes in massive galaxies are missed because their broad emission lines are swamped by the host galaxy's narrow components or fall below detection thresholds.
• Inconsistent Results: Previous analyses have yielded conflicting conclusions. Some suggest SMBHs at high redshift are systematically "overmassive" compared to local relations, while others argue this is an artifact of selection effects or that the relation is consistent with local normalizations but with high scatter.
• Methodological Gaps: Many studies lack a rigorous, forward-modeling approach to explicitly account for the complex, multivariate detectability of broad H$\alpha$ lines, which depends on black hole mass, stellar mass, star formation rate (SFR), and Eddington ratio ($\lambda_{Edd}$).

2. Methodology

The authors developed a forward-modeling Bayesian framework designed to explicitly incorporate selection effects into the likelihood function. The methodology consists of four main components:

A. Synthetic Data Generation (Mock Spectra)

• Physical Inputs: The authors generated mock H$\alpha$ spectra based on four physical parameters: Black Hole Mass ($M_{BH}$), Stellar Mass ($M_\star$), Eddington Ratio ($\lambda_{Edd}$), and Star Formation Rate (SFR).
• Spectral Modeling:
  • Narrow Component: Modeled as a Gaussian with luminosity scaling linearly with SFR and width derived from the host galaxy's dynamical mass (using the $M_\star$–$R_{eff}$ relation).
  • Broad Component: Modeled as a Gaussian with luminosity derived from the 5100 Å continuum and width determined by virial relations linking $M_{BH}$, luminosity, and Full Width at Half Maximum (FWHM).
• Observational Conditions: Spectra were convolved with the NIRSpec line spread function ($\Delta\lambda/\lambda \approx 2700$) and noise was added. Two noise levels were tested: a deep JADES-like depth ($6 \times 10^{-21}$erg s$^{-1}$cm$^{-2}$Å$^{-1}$) and a shallower, more conservative depth ($2.4 \times 10^{-20}$erg s$^{-1}$cm$^{-2}$Å$^{-1}$).

B. Detectability Mapping

• Fitting Criteria: Mock spectra were fitted with single-Gaussian (narrow only) and double-Gaussian (narrow + broad) models. A broad component was deemed detected if:
  1. $\Delta$BIC (Bayesian Information Criterion) $> 10$ (strong statistical preference for the double model).
  2. The single-Gaussian width $\sigma_{single} > 2\sigma_{dyn}$ (ensuring broadening is not due to host kinematics).
• Selection Function: By sampling a 4D grid of physical parameters ($M_\star, M_{BH}, \lambda_{Edd}, \text{SFR}$), the authors computed the detection probability ($P_{det}$) for every point in the $(\log M_\star, \log M_{BH})$ plane. This generated 2D detectability maps showing where broad lines are recoverable under different noise regimes.

C. Truncated-Likelihood Bayesian Inference

• Model: The intrinsic $M_{BH}–M_\star$ relation is modeled as a log-linear scaling with orthogonal intrinsic scatter: $\log M_{BH} = \alpha + \beta \log M_\star$.
• Likelihood Construction: The authors used a truncated likelihood approach. The detectability maps derived in step B define a "visibility boundary" in parameter space. The likelihood function is truncated below this boundary to correct for the fact that sources below the threshold are unobservable.
• Parameters: The model solves for the normalization ($\alpha$), slope ($\beta$), and intrinsic orthogonal scatter ($\sigma_{int}$), while marginalizing over observational uncertainties (fixed at 0.4 dex for both masses).

D. Validation

• Recovery Tests: The pipeline was validated by generating mock samples from known local relations (Kormendy & Ho 2013; Reines & Volonteri 2015), applying the detectability cuts, and verifying that the truncated-likelihood method successfully recovered the input parameters within 68% credible intervals.

3. Key Results

Applying this framework to the homogeneous broad-line AGN sample from the JADES survey (compiled by Juodžbalis et al. 2025a), the authors derived the following constraints for $z \gtrsim 4$:

• Scaling Relation:
  $$\log M_{BH} = -4.06^{+0.50}_{-0.51} + 1.17^{+0.06}_{-0.06} \log M_\star$$
• Intrinsic Scatter:
  $$\sigma_{int} = 0.63^{+0.14}_{-0.11} \text{ dex}$$
• Comparison with Local Relations:
  • The slope and normalization are consistent with the local relation determined by Kormendy & Ho (2013).
  • The relation is significantly steeper and higher than the lower-normalization relation of Reines & Volonteri (2015).
  • The result lies slightly below the high-redshift trend proposed by Pacucci et al. (2023), suggesting the "overmassive" black hole claim is likely driven by selection bias rather than a true shift in the mean relation.
• The Role of Scatter: The most significant finding is the substantially larger intrinsic scatter ($\sigma_{int} \approx 0.63$) compared to the local Universe ($\sim 0.3$ dex). This indicates that the evolution of the $M_{BH}–M_\star$ relation at high redshift is driven by an increase in dispersion rather than a systematic shift in the mean.

4. Significance and Physical Implications

• Co-evolution Timeline: The consistency of the mean relation with local determinations suggests that the average $M_{BH}–M_\star$ scaling was already established by $z \sim 4–6$.
• Diversity of Growth Histories: The large intrinsic scatter implies a wide diversity in black hole–galaxy growth histories at early cosmic times. This is likely driven by:
  • Bursty Accretion: Stochastic, intermittent gas accretion episodes leading to temporary over- or under-massive states.
  • Delayed Feedback: Feedback mechanisms may not yet be fully self-regulating.
  • Seeding and Mergers: Variations in initial seed masses and a lack of sufficient hierarchical merging events to "wash out" initial differences (statistical averaging).
• Methodological Impact: The study demonstrates that apparent evolutionary trends (like overmassive black holes) can be artifacts of incomplete selection modeling. By explicitly forward-modeling detectability, the authors recover a relation consistent with local physics, resolving previous tensions in the literature.
• Future Outlook: The results highlight the need for larger, deeper JWST samples and improved stellar mass estimates to further constrain the evolution of the scatter and map the diversity of SMBH growth in the early Universe.

Conclusion

This paper provides a statistically rigorous, selection-aware determination of the $M_{BH}–M_\star$ relation at $z \gtrsim 4$. It concludes that while the mean scaling relation is consistent with local Universe measurements (specifically Kormendy & Ho 2013), the scatter is significantly enhanced. This suggests that in the early Universe, SMBHs and galaxies were co-evolving through more stochastic and diverse pathways than in the settled local Universe.