🔭 astrophysics

Cosmological Evolution of Gamma Ray Bursts

This paper investigates the cosmological evolution of long gamma-ray bursts using an expanded dataset of measured and machine learning-estimated redshifts, finding that while a low-redshift excess persists in the measured sample, incorporating ML-estimated redshifts reveals a mid-range overabundance that reduces this excess, challenging the traditional separation of burst progenitors.

Original authors: Sujay Champati, Vahé Petrosian, Maria G. Dainotti

Published 2026-06-19

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Sujay Champati, Vahé Petrosian, Maria G. Dainotti

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the universe as a giant, bustling city that has been growing and changing for billions of years. In this city, Gamma-Ray Bursts (GRBs) are like massive, blinding fireworks that go off in the sky. They are the brightest explosions in the cosmos, visible from incredibly far away.

For a long time, astronomers thought these fireworks came from two different types of "fireworks factories":

Long Bursts (LGRBs): These were thought to be caused by massive stars collapsing (like a building imploding). Since new stars are born when the city is busy building, these bursts should follow the city's "construction schedule" (the rate of star formation).
Short Bursts (SGRBs): These were thought to be caused by two dense objects crashing into each other (like two cars colliding). These happen later, after the objects have had time to drift apart and meet up.

The Mystery
Recently, astronomers noticed something weird. When they looked at the "construction schedule" (how many stars are being born) and compared it to where the Long Bursts were happening, the math didn't add up. Specifically, there were way too many Long Bursts happening in the "neighborhoods" close to us (low redshift) compared to what the construction schedule predicted. It was like finding a sudden, unexplained spike in fireworks in a quiet suburb.

Some recent discoveries even showed that a few of these "Long" bursts were actually linked to the "Short" burst factories (kilonovae), suggesting the two types of factories might be more similar than we thought.

The New Investigation
The authors of this paper wanted to solve this mystery. The problem with past studies was that they were like trying to guess the population of a whole country by counting only 200 people. The sample size was too small to be sure.

So, they gathered a much bigger group of data:

They took all the known Long Bursts with confirmed distances (redshifts).
They added a huge new batch of bursts where the distance was guessed using Machine Learning (AI). This AI had been trained to look at the "fingerprint" of the burst and predict how far away it is.

The Tools They Used
To make sense of this data, they had to deal with a tricky problem: The "Spotlight Bias."
Imagine you are looking at a dark room with a flashlight. You can easily see the bright people standing right in front of you, but you might miss the dim people standing far away, even if there are many of them. In astronomy, we are biased toward seeing bright bursts nearby and missing faint ones far away.

To fix this, the authors used a special statistical method (called the Efron-Petrosian method) that acts like a "correction lens." It mathematically removes the bias of the flashlight, allowing them to see the true distribution of bursts as if they were all equally visible.

What They Found

The "Local" Excess is Real: When they looked at the confirmed data (without the AI guesses), they confirmed the mystery: there is indeed a huge excess of Long Bursts happening in the nearby universe. The rate of these bursts is much higher than the rate of new stars being born in that same area.
The AI Smooths Things Out: When they added the AI-predicted bursts to the mix, the picture changed slightly. The AI data had a lot of bursts in the "middle-distance" neighborhoods. Adding these in made the "local excess" look a little less extreme, though it didn't disappear completely.
The Connection to Stars: For the distant universe (far away), the bursts follow the star-formation schedule perfectly. But for the nearby universe, something else is going on.

The Conclusion
The paper suggests that the "extra" bursts we see nearby might not be coming from the usual collapsing stars. Because of recent discoveries linking some bursts to crashing neutron stars (kilonovae), the authors propose that many of these nearby Long Bursts might actually be caused by crashing objects, not just collapsing stars.

This is a big deal because it changes our understanding of what causes these explosions and suggests there might be more sources of gravitational waves (ripples in space-time) in our local neighborhood than we previously thought.

In short: The universe is throwing more "Long" fireworks near us than expected. By using a bigger dataset and a smart AI to fill in the gaps, the authors confirmed this spike exists and suspect it's caused by a different type of cosmic crash than we originally believed.

Technical Summary: Cosmological Evolution of Gamma Ray Bursts

Problem Statement
Gamma-ray bursts (GRBs) are traditionally classified into long-duration (LGRBs) and short-duration (SGRBs) events, associated respectively with massive star collapse (collapsars) and compact object mergers. Theoretical expectations suggest LGRBs should trace the cosmic star formation rate (SFR), while SGRBs follow a delayed SFR. However, recent observational evidence, including the association of specific LGRBs (e.g., GRB211211A, GRB230307A) with kilonovae, challenges this strict dichotomy. Furthermore, multiple studies have reported an excess of LGRBs at low redshifts compared to SFR predictions. Previous analyses of this discrepancy have been limited by small sample sizes (typically <200 spectroscopic redshifts) and potential selection biases. This paper aims to reassess the luminosity evolution and formation rate of LGRBs using a significantly larger dataset, incorporating both spectroscopic redshifts and machine learning (ML)-estimated redshifts, to determine if the low-redshift excess is a physical phenomenon or an artifact of sample incompleteness.

Methodology
The study utilizes a comprehensive dataset comprising:

Spectroscopic Sample: 444 LGRBs with measured redshifts from the Swift observatory (after filtering for upper/lower bounds).
ML-Augmented Sample: 251 LGRBs with redshifts estimated via a machine learning method (Narendra et al. 2025), filtered to exclude $z > 14.4$ .
Total Sample: A combined catalog of 695 bursts, with 499 possessing the necessary flux and spectral data (from Swift/BAT and Fermi/GBM catalogs) to calculate rest-frame luminosities.

To address observational selection biases (Malmquist/Eddington bias) inherent in flux-limited surveys, the authors employ non-parametric, non-binning techniques:

Efron-Petrosian (EP) Method: Used to test for correlations between luminosity ( $L$ ) and redshift ( $Z$ ) under one-sided truncation. This allows for the determination of a luminosity evolution function, $g(Z)$ , such that the de-evolved luminosity $L_0 = L/g(Z)$ is uncorrelated with redshift.
Lynden-Bell $C^-$ Method: Applied to the de-evolved data to reconstruct the bias-corrected cumulative luminosity function ( $\phi(L_0)$ ) and the cumulative source co-moving formation rate ( $\dot{\sigma}(Z)$ ).

The analysis involves calculating K-corrections using a running average of spectral fits (Power Law and CPL models) to define a conservative truncation boundary. The robustness of the results is tested against perturbations in the average K-correction and the luminosity evolution parameter $k$ .

Key Results

Luminosity Evolution: A strong correlation between luminosity and redshift is confirmed for both the full and spectroscopic-only catalogs. The EP method yields a luminosity evolution scaling of $L \sim Z^k$ , with $k \approx 2.8$ for the full catalog and $k \approx 3.7$ for the spectroscopic-only catalog.
Luminosity Function: The local (de-evolved) luminosity function is well-described by a smoothly broken power law. The full catalog exhibits a higher normalization ( $\phi_0 \approx 3.08 \times 10^2$ ) compared to the spectroscopic-only sample ( $\phi_0 \approx 1.72 \times 10^2$ ), reflecting the increased sample size.
Formation Rate and Low-Redshift Excess:
- Spectroscopic Catalog: The derived formation rate density ( $\dot{\rho}$ ) tracks the cosmic SFR for $z \gtrsim 1.5$ . However, for $z < 1.5$ , the LGRB rate deviates significantly, rising to approximately two orders of magnitude above the SFR at the lowest redshifts.
- Full (ML-Augmented) Catalog: The inclusion of ML-estimated redshifts reduces the magnitude of this low-redshift excess. The formation rate remains consistent with the SFR down to $z \sim 1$ , after which it breaks and increases, but by a smaller factor than in the spectroscopic-only case.
- Redshift Distribution: The ML sample is concentrated in the mid-range redshifts ( $1.5 < z < 3$ ) and contributes relatively few low-redshift sources. The authors note that the ML-predicted redshifts in the small overlap sample tend to be lower than spectroscopic values, suggesting the reduction in the low-redshift excess is not due to a systematic upward shift in ML redshifts, but rather the specific distribution of the added ML sources.

Significance and Claims
The paper claims to provide updated constraints on LGRB progenitors and their evolutionary history by utilizing the largest available sample of LGRBs with redshifts to date. The primary significance lies in:

Validation of the Low-Redshift Excess: The study confirms the existence of a low-redshift excess in LGRB formation rates relative to the SFR using a large, spectroscopic sample, consistent with previous smaller studies.
Impact of ML Data: It demonstrates that incorporating ML-estimated redshifts alters the inferred formation history, specifically by reducing the apparent magnitude of the low-redshift excess. This highlights the sensitivity of rate evolution studies to the redshift distribution of the sample.
Progenitor Implications: The persistence of the low-redshift excess, even after accounting for sample size and ML augmentation, supports the hypothesis that a subset of LGRBs may originate from compact mergers (NS-NS or NS-BH), as suggested by recent kilonova associations. The authors suggest this could increase the expected rate of gravitational wave sources.
Methodological Robustness: The study confirms that the inferred low-redshift enhancement is stable across the $1\sigma$ uncertainty range of the luminosity evolution parameter $k$ , indicating the result is not an artifact of the specific evolution correction chosen.

The authors conclude that while the low-redshift excess is a robust feature of the spectroscopic data, its precise magnitude and physical origin (e.g., sample incompleteness vs. genuine physical excess) require further investigation, particularly as the field moves toward larger, ML-augmented datasets.

Technical Summary: Cosmological Evolution of Gamma Ray Bursts

More like this