Galactic bars and active galactic nucleus fuelling in the second half of cosmic history

This study utilizes deep learning to analyze HSC images up to $z\sim 0.8$ and finds that galactic bars contribute to fuelling low-to-moderate luminosity Active Galactic Nuclei, whereas major mergers remain the primary driver for the most powerful accretion events.

The Gist

Imagine the universe as a vast, bustling city of galaxies. At the center of the most massive buildings in this city sits a "supermassive black hole." Usually, these black holes are like sleeping giants, but sometimes they wake up, start eating, and glow incredibly bright. When they do, we call them Active Galactic Nuclei (AGN).

For a long time, astronomers thought the only way to wake these giants up was a massive, violent crash between two galaxies—a "major merger." It's like thinking a sleeping dragon only wakes up if you drop a boulder on it.

But this new paper asks a different question: Can the galaxy wake up the dragon all by itself, without a crash? Specifically, can a "bar" inside the galaxy do the job?

The Galactic Bar: The Cosmic Conveyor Belt

Think of a spiral galaxy (like our Milky Way) as a giant, spinning pizza dough. Sometimes, the dough doesn't just spin; it stretches out into a long, straight line in the middle. Astronomers call this a bar.

Imagine this bar as a conveyor belt or a funnel. Its job is to grab gas and dust from the outer edges of the galaxy and slide it straight toward the center, right into the mouth of the black hole. If the conveyor belt works, the black hole gets fed, wakes up, and starts glowing.

The Big Experiment

The authors of this paper wanted to test if this "conveyor belt" theory is true. They looked at thousands of galaxies from the last 6 billion years of the universe's history (the "second half of cosmic history").

To do this, they used a clever trick:

1. The Eyes: They used a powerful telescope (the Subaru Telescope) to take high-definition photos of these galaxies.
2. The AI Brain: They trained a Deep Learning computer model (named Zoobot) to look at these photos and spot the bars. It's like teaching a computer to play "I Spy" with galaxies, distinguishing between those with a clear bar, a weak bar, and no bar at all.
3. The Match: They compared galaxies with bars to galaxies without bars, making sure the two groups were identical in size, age, and color, so the only difference was the presence of the bar.

The Findings: It's Complicated!

The results were surprising and nuanced. Here is the breakdown in simple terms:

1. Bars do help, but only for "snacks."
The study found that galaxies with bars do have more active black holes than those without. However, this effect is mostly seen in low-to-moderate power black holes.

• Analogy: Think of the bar as a reliable delivery driver. It's great at bringing a steady stream of groceries (gas) to the kitchen (the black hole) to make a nice dinner (a moderate AGN). It keeps the black hole active and happy.

2. The "Super-Feast" requires a crash.
Here is the twist: When the black hole becomes a monster, glowing with extreme power (the most luminous AGN), the bars disappear or stop working.

• The Discovery: The team found almost zero galaxies with bars hosting the most powerful, dominant black holes.
• Analogy: If the black hole needs a "super-feast" to become a quasar (the brightest type of AGN), the conveyor belt isn't enough. It takes a galaxy crash (a major merger) to dump a massive pile of fuel on the fire all at once.
• Furthermore, the paper suggests that when a black hole gets too powerful, the energy it releases might actually break the conveyor belt (destroy the bar) or stop it from working. It's like the dragon breathing fire so hot it melts the conveyor belt feeding it.

The Conclusion: Two Ways to Wake the Dragon

The paper helps solve a long-standing debate. It suggests there are two distinct ways to feed a supermassive black hole:

1. The Secular Way (The Bar): Slow, steady, and internal. The galaxy's own structure (the bar) funnels gas to the center. This powers the "everyday" active black holes we see often.
2. The Violent Way (The Merger): Fast, chaotic, and external. Two galaxies smash together, dumping huge amounts of gas into the center. This powers the rare, super-bright, "monster" black holes.

In a nutshell: Bars are the reliable delivery trucks that keep the black holes fed for a steady meal. But if you want to throw a massive, universe-shaking party, you need a galaxy crash. The conveyor belt can't handle the VIPs.

Technical Summary

Here is a detailed technical summary of the paper "Galactic bars and active galactic nucleus fuelling in the second half of cosmic history" by La Marca et al. (2026).

1. Problem Statement

The mechanisms driving the growth of Supermassive Black Holes (SMBHs) and the triggering of Active Galactic Nuclei (AGN) remain a central topic in galaxy evolution. While major mergers are widely accepted as the primary trigger for the most luminous AGN (quasars), the role of secular processes, specifically galactic bars, in fuelling lower-luminosity AGN is debated.

• The Conflict: Some studies suggest bars efficiently funnel gas to the central SMBH, increasing AGN incidence. Others find no correlation or argue that differences vanish when controlling for host galaxy properties (stellar mass, color).
• The Gap: Previous studies often struggle with sample biases, diverse AGN selection methods, and the difficulty of distinguishing between merger-induced and bar-induced activity, particularly in the "second half of cosmic history" ($z \sim 0.1 - 0.8$).
• Objective: This paper aims to quantify the specific contribution of galactic bars to AGN fuelling by isolating disc-dominated, non-merging systems and analyzing the relationship between bar presence and AGN properties (both relative and absolute luminosity).

2. Methodology

Data and Sample Selection

• Parent Sample: Utilized the galaxy sample from La Marca et al. (2024, LM24), drawn from the KiDS-N-W2 field ($\sim 65$ deg$^2$) with multi-wavelength coverage (X-ray, optical/NIR, MIR, Far-IR).
• Redshift & Mass: Selected galaxies in the range $0.1 \le z \le 0.76$with stellar mass limits ensuring completeness ($M_\star \ge 10^9 M_\odot$or$2.5 \times 10^9 M_\odot$ depending on redshift).
• Exclusion: Major mergers were explicitly excluded from the primary analysis to isolate secular processes. Their impact was tested separately in Section 5.
• AGN Identification: AGN were selected using three independent diagnostics to ensure robustness:
  1. Mid-Infrared (MIR) Colors: $W1 - W2 > 0.8$ mag (Stern et al. 2012).
  2. X-ray Detections: Cross-matched with the eFEDS catalogue.
  3. Spectral Energy Distribution (SED) Fitting: Using CIGALE (Code Investigating GALaxy Emission). AGN were defined as having an AGN fraction $f_{\text{AGN}} \ge 0.1$ (in the 3–30 $\mu$m range), a stricter threshold than previous work to ensure robust AGN contribution.

Bar Identification via Deep Learning

• Imaging Data: Hyper Suprime-Cam (HSC-SSP) $i$-band images.
• Model: A Convolutional Neural Network (CNN) based on the Zoobot framework (Walmsley et al. 2023), specifically the zoobot-encoder-convnext_nano architecture.
• Training: Fine-tuned on a balanced dataset of 21,847 galaxies classified by Galaxy Zoo volunteers. The training set included five classes: Strongly Barred, Weakly Barred, Unbarred Discs, Smooth Galaxies, and Edge-on Galaxies.
• Classification Criteria:
  • Barred ($A_{\text{bar}}$): Sum of probabilities for strong/weak bars $\ge 0.65$.
  • Strongly Barred ($S_{\text{bar}}$): $P_{\text{strong}} \ge P_{\text{weak}}$.
  • Weakly Barred ($W_{\text{bar}}$): $P_{\text{strong}} < P_{\text{weak}}$.
  • Unbarred Disc ($U_{\text{bar}}$): $P_{\text{unbarred}} > 0.45$ and significantly higher than the second-highest probability ($\Delta P > 0.1$).

Control Sample Construction

To mitigate biases from host galaxy properties, a matched control sample was constructed for every barred galaxy. Unbarred controls were matched to barred galaxies in:

• Redshift ($z$)
• Stellar Mass ($M_\star$)
• Rest-frame color ($g-r$)
• Constraint: At least 10 unique controls per barred galaxy were required.

3. Key Contributions

1. Deep Learning Morphology: Application of a fine-tuned Zoobot model to HSC-SSP data to robustly classify weak and strong bars in a large sample up to $z \sim 0.8$, overcoming limitations of visual inspection at higher redshifts.
2. Multi-Diagnostic AGN Selection: Simultaneous use of MIR, X-ray, and SED-derived AGN properties to cross-validate results and capture different AGN populations.
3. Continuous Analysis: Moving beyond binary AGN presence/absence to analyze the AGN fraction ($f_{\text{AGN}}$) and accretion disc luminosity ($L_{\text{disc}}$) as continuous variables.
4. Merger Isolation: Explicitly separating the effects of secular bars from major mergers, providing a clearer picture of secular fuelling mechanisms.

4. Key Results

AGN Incidence (Binary Classification)

• Enhanced Fraction: Barred galaxies host a higher fraction of AGN compared to unbarred controls across all selection methods.
  • MIR AGN: $\sim 6\times$ excess (statistically limited by small numbers).
  • X-ray AGN: $\sim 2.7\times$ excess (statistically robust).
  • SED AGN: $\sim 1.12\times$ excess (modest but statistically significant, $p \approx 0.027$).
• Conclusion: Bars contribute to the global AGN budget, particularly for X-ray and SED-selected populations.

Relative AGN Power ($f_{\text{AGN}}$)

• Low-to-Moderate Dominance: The fraction of barred galaxies ($f_{\text{bar}}$) remains relatively flat ($\sim 20-50\%$) for AGN with $f_{\text{AGN}} < 0.75$.
• The "Deaf" Zone: A significant dearth of barred galaxies is observed for AGN with $f_{\text{AGN}} > 0.75$.
  • Galaxies with dominant AGN emission ($f_{\text{AGN}} > 0.8$) are almost exclusively unbarred and often show merger features.
  • This contrasts sharply with the merger fraction ($f_{\text{merg}}$), which rises steeply for $f_{\text{AGN}} > 0.8$.

Absolute AGN Luminosity ($L_{\text{disc}}$)

• Inverse Correlation at High Luminosity: The fraction of barred galaxies decreases as AGN luminosity increases.
  • Bars are common in AGN with $L_{\text{disc}} \approx 10^{43} - 10^{44.5}$ erg s$^{-1}$.
  • For the most luminous AGN ($L_{\text{disc}} > 10^{45}$ erg s$^{-1}$), the bar fraction drops to near zero.
• Comparison with Mergers: While the merger fraction increases with luminosity, the bar fraction decreases, suggesting a dichotomy in fuelling mechanisms.

Robustness Checks

• Re-including major mergers in the analysis (Section 5) did not alter the primary conclusions: the scarcity of bars in high-$f_{\text{AGN}}$ and high-$L_{\text{disc}}$ systems remains a genuine feature, not an artifact of sample selection.

5. Significance and Conclusion

The paper resolves the long-standing debate on bar-driven AGN fuelling by defining its operational boundaries:

1. Secular Fuelling: Galactic bars are effective triggers for low-to-moderate luminosity AGN (Seyfert-like activity). They contribute to the global AGN population but are not the dominant mechanism for the most powerful events.
2. Merger Dominance: Major mergers are the principal (and likely exclusive) mechanism for triggering the most luminous and AGN-dominated systems ($f_{\text{AGN}} > 0.75$, $L_{\text{disc}} > 10^{45}$ erg s$^{-1}$).
3. Physical Interpretation: The lack of bars in high-power AGN suggests two possibilities:
   • Bars cannot channel gas at rates sufficient to power the most extreme accretion events.
   • The intense feedback or dynamical disruption from a dominant AGN (or the merger that triggered it) destroys the bar structure.

Future Outlook: The authors suggest that future data from the Euclid mission will provide the statistical power necessary to extend these findings to higher redshifts and refine the understanding of bar evolution and AGN co-evolution across cosmic time.