🔭 astrophysics

The JADES Transient Survey II: Volumetric Supernova Rates out to z~5

Using the JADES Transient Survey's identification of 83 supernova candidates in deep JWST fields, this study derives the first volumetric core-collapse and Type Ia supernova rates out to redshift $z \sim 5$ , confirming consistency with cosmic star formation expectations and providing the first direct observational evidence of a decline in core-collapse rates beyond cosmic noon.

Original authors: Christa DeCoursey, Christian Vassallo, Louis-Gregory Strolger, Justin D. R. Pierel, Eiichi Egami, Seppo Mattila, Armin Rest, David A. Coulter, Andrew J. Bunker, Alex J. Cameron, James M. DerKacy, Dani

Published 2026-07-15

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Christa DeCoursey, Christian Vassallo, Louis-Gregory Strolger, Justin D. R. Pierel, Eiichi Egami, Seppo Mattila, Armin Rest, David A. Coulter, Andrew J. Bunker, Alex J. Cameron, James M. DerKacy, Daniel J. Eisenstein, Michael Engesser, Ori D. Fox, Sebastian Gomez, Massimo Griggio, Kevin Hainline, Ryan Hausen, Zhiyuan Ji, Benjamin D. Johnson, Roberto Maiolino, Takashi J. Moriya, Brant Robertson, Koji Shukawa, Matthew R. Siebert, Fengwu Sun, Sandro Tacchella, Christina C. Williams, Christopher N. A. Willmer, Yossef Zenati

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 construction site. For decades, astronomers have been trying to count how many massive stars are being born and then dying in spectacular explosions called supernovae. They've been able to count these explosions easily in our "neighborhood" of the universe, but as they look further back in time—toward the "cosmic noon" when the universe was about half its current age—the construction site gets foggy, and the explosions get harder to spot.

Enter the JADES Transient Survey, a team of astronomers using the James Webb Space Telescope (JWST) as a super-powered flashlight to peer into this foggy, ancient construction zone. Their mission? To count the explosions of dying massive stars (called Core-Collapse Supernovae) and the explosions of white dwarf stars (called Type Ia Supernovae) in a region of space so deep and dark it's like looking at a single grain of sand from a mile away.

The Big Discovery: Counting the Cosmic Fireworks

The team found 83 potential supernova candidates in a tiny patch of sky. From this group, they calculated the rate at which these explosions happen in a specific volume of space. Here is what they found, broken down by the type of explosion:

1. The Massive Star Explosions (Core-Collapse Supernovae)
These are the stars that run out of fuel and collapse under their own weight. The team measured how often these happen between redshifts z ∼2 and z ∼5 (a time when the universe was roughly 2 to 12 billion years ago).

The Numbers: In the time period between 2.06 ≤z < 2.78, they found a rate of 6.2 +2.2 −1.7 (in units of 10⁻⁴ CC SNe yr⁻¹ Mpc⁻³). In the even older, more distant period of 2.78 ≤z ≤5.06, the rate was 4.1 +1.5 −1.1.
The Trend: The data suggests that after a peak around "cosmic noon" (when star formation was highest), the number of these explosions starts to drop off. This is the first direct observational hint that the rate of massive star deaths follows the same decline as the rate of star births in the early universe. Before this, it was just a guess based on other data; now, we have a direct look.

2. The White Dwarf Explosions (Type Ia Supernovae)
These are different; they happen when a white dwarf star steals too much material from a partner and blows up.

The Numbers: For the period 1.92 ≤z < 3.60, they measured a rate of 0.3 +0.3 −0.2 × 10⁻⁴ SNe Ia yr⁻¹ Mpc⁻³.
The Trend: The data is too sparse to say for sure if these rates go up or down, but they appear to stay relatively flat across the time they observed.

The "One-Shot" Challenge: Guessing with a Single Snapshot

Here is the tricky part of the story. Usually, to identify a supernova, astronomers watch it over time, like watching a movie to see the plot unfold. But in this deep, distant survey, most of the 83 candidates were only caught once. It's like trying to identify a person at a party by seeing them for just one second.

To make sure their "one-shot" guesses were accurate, the team ran a massive simulation. They created ~23,000 fake supernova light curves (mock SEDs) and tried to classify them using the same single-shot method.

The Result: The computer was pretty good at telling the difference between a massive star explosion and a white dwarf explosion (getting it right about 87% of the time for massive stars and 45–52% for white dwarfs, depending on the data quality).
The Catch: The computer struggled to tell the difference between the subtypes of massive star explosions (like distinguishing a "Type IIP" from a "Type Ib/c"). It was like the computer could tell the difference between a dog and a cat, but couldn't tell a Golden Retriever from a Poodle. Because of this, they couldn't give rates for specific subtypes, only for the broad categories.

Two Lists, One Story

To be extra careful, the team presented their results in two ways:

The "Full" Sample: This includes all 83 candidates, even the ones they only saw once.
The "Gold" Sample: This is a smaller, stricter list containing only the supernovae they could identify with high confidence (either by seeing them multiple times or by taking a "spectral fingerprint" with a spectrograph).

Both lists told the same story: the rates are consistent with what we expect based on how many stars were being born at that time. The "Gold" sample is more reliable but has bigger error bars because it's so small. The "Full" sample is larger but carries a tiny risk that a few misidentified explosions might be sneaking in.

What They Didn't Find (and What They Ruled Out)

The paper is very clear about what it doesn't claim:

No "Exotic" Rates: They did not find evidence that the rules of physics change for these stars in the early universe. The rates match the predictions based on star formation, suggesting that massive stars were dying just as we thought they would.
No Subtype Breakdown: They explicitly ruled out the ability to calculate rates for specific subtypes of massive star explosions (like IIP vs. Ib/c) using only single-epoch data. The confusion between these types was too high to trust the numbers.
No "Solved" Problem: They emphasize that while they see a decline in rates at the highest redshifts, the error bars are still large. They are suggesting a trend, not proving it with absolute certainty. They also note that their rates might be slightly lower than expected because their method for calculating how "visible" the stars were might have been a bit too optimistic.

The Bottom Line

This paper is a pioneering step. It's the first time we've directly measured how often massive stars explode in the very early universe, using the most powerful telescope ever built. While the numbers come with big "maybe" signs (the error bars are wide), the data suggests that the universe's history of star deaths is tightly linked to its history of star births, even billions of years ago.

As the authors note, future surveys with JWST and the upcoming Roman Space Telescope will catch more of these cosmic fireworks, shrinking those error bars and turning these "suggestions" into solid facts. For now, we have our first real glimpse of the universe's explosive youth.

Technical Summary: The JADES Transient Survey II: Volumetric Supernova Rates out to z ∼5

Problem Statement
While volumetric supernova (SN) rates have been extensively measured at low redshifts ( $z < 1$ ) and tentatively out to $z \sim 2.5$ , a gap remains in direct observational constraints for the high-redshift Universe ( $z \sim 2–5$ ). Core-collapse (CC) SNe are expected to trace the instantaneous cosmic star formation rate density (SFRD), which peaks at "cosmic noon" ( $z \sim 2$ ) and declines thereafter. However, previous measurements at $z > 2$ have been limited by small sample sizes, large statistical uncertainties, and a lack of spectroscopic confirmation. Similarly, Type Ia (SN Ia) rates at $z > 1$ remain highly uncertain, hindering the construction of robust delay-time distributions (DTDs) necessary to distinguish between single-degenerate and double-degenerate progenitor scenarios. The James Webb Space Telescope (JWST) offers the sensitivity and spatial resolution required to push the transient redshift frontier to $z \sim 5$ , but classifying these distant events is challenging due to sparse light-curve sampling (often single-epoch) and the difficulty of distinguishing SN types photometrically without phase constraints.

Methodology
This study utilizes the JADES Transient Survey (JTS) sample, which identified 83 SN candidates in the $\sim 25$ arcmin $^2$ JADES Deep Field using multi-epoch JWST/NIRCam imaging. The analysis proceeds through three primary methodological stages:

Classification Scheme and Validation: The authors reclassified the JTS sample using the STARDUST2 Bayesian light-curve fitting tool, employing a modified scheme that restricts SN Ia stretch ( $x_1$ ) and color ( $c$ ) parameters to physical bounds and fixes $x_1$ for single-epoch sources. To quantify the accuracy of photometric classification based on single spectral energy distributions (SEDs), the team generated $\sim 23,000$ mock SN SEDs across various subtypes (SN Ia, Ib/c, IIP, IIL, IIn), redshifts ( $0.7 \le z \le 5$ ), phases, and extinction levels. These mocks were passed through the classifier to construct confusion matrices, revealing that while the classifier effectively distinguishes CC SNe from SNe Ia (average true positive rate $\sim 0.87$ for CC, $\sim 0.45–0.52$ for Ia), it struggles to differentiate between CC subtypes (e.g., IIP vs. Ib/c) and often misclassifies abnormal SNe Ia.
Volumetric Rate Calculation: The authors employ a probabilistic "control time" formalism to calculate volumetric rates. This method accounts for survey cadence, depth, area, classification uncertainty, and redshift uncertainty.
- Visibility Window: Synthetic light curves (using nugent models for CC SNe and nugent-sn1a for SNe Ia) are integrated over detection efficiency curves to determine the time an SN remains detectable.
- Control Time: The effective detectability window is folded with the survey cadence (two main epochs separated by one observer-frame year).
- Rate Derivation: Two samples are analyzed: (1) the "full sample" including all photometrically classified sources (including single-SED classifications), and (2) a "gold sample" restricted to sources with spectroscopic confirmation or multi-epoch light curves. Rates are computed using Equation 5 (for CC SNe assuming intrinsic subtype fractions) and Equation 15 (for SNe Ia), integrating over comoving volume and correcting for detection probabilities.
Correction Attempts: The authors attempted to apply Bayesian and frequentist correction factors to account for misclassification in the single-SED sample. However, these methods were deemed ineffective: the Bayesian approach yielded no change for sources with extreme probability outputs ($0$ or $1$), and the frequentist approach resulted in uncertainties too large to provide meaningful constraints. Consequently, the final rates are presented uncorrected for single-SED misclassification, with appropriate caveats.

Key Results

Core-Collapse SN Rates: The study reports the first volumetric CC SN rates out to $z \sim 5$ .
- Full Sample Rates: In units of $10^{-4}$ CC SNe yr $^{-1}$ Mpc $^{-3}$ , the rates are $6.2^{+2.2}_{-1.7}$ for $2.06 \le z < 2.78$ and $4.1^{+1.5}_{-1.1}$ for $2.78 \le z \le 5.06$ .
- Trends: The rates are broadly consistent with expectations derived from galaxy luminosity-based measurements of the cosmic SFRD (Madau & Dickinson 2014; Harikane et al. 2022). The full sample tentatively exhibits the predicted decline in rates beyond cosmic noon ( $z > 2$ ), providing the first direct observational indication of this behavior, though large uncertainties prevent definitive conclusions.
- Gold Sample: The restricted "gold" sample shows rates consistent with the full sample within $1\text{--}2\sigma$ , though with larger error bars due to smaller sample sizes.
Type Ia SN Rates:
- The full sample yields a rate of $0.3^{+0.3}_{-0.2} \times 10^{-4}$ SNe Ia yr $^{-1}$ Mpc $^{-3}$ at $1.92 \le z < 3.60$ .
- The "gold" sample contains only one redshift bin ( $z = 2.90–4.17$ ) due to the scarcity of high- $z$ SNe Ia with spectroscopic or multi-epoch confirmation.
- The combined data suggest flat SN Ia rates from $z \sim 1$ to $\sim 6$ , though the highest-redshift bin is shown as translucent to indicate potential CC SN contamination.

Significance and Claims
The paper claims to provide the first direct volumetric measurements of both CC SN and SN Ia rates in the redshift range $z \sim 2–5$ . Its primary significance lies in:

Extending the Redshift Frontier: Pushing SN rate measurements beyond the previous HST frontier of $z \sim 2.5$ to $z \sim 5$ using JWST.
Validating Photometric Classification: Demonstrating that while single-SED photometric classification is imperfect (particularly for distinguishing SN Ia from CC SNe and among CC subtypes), it yields rates consistent with the more robust "gold" sample, suggesting that large high- $z$ surveys can rely on photometric classifications with appropriate statistical handling.
Observational Evidence of SFRD Decline: Offering the first direct observational hint that CC SN rates decline beyond cosmic noon, aligning with the decline in cosmic star formation density.
Foundation for Future Work: Establishing a baseline for future high- $z$ SN surveys (e.g., with the Roman Space Telescope) to provide more robust constraints on SN rates and progenitor physics.

The authors remain modest regarding their claims, noting that the observed rates are consistently lower than SFRD-based expectations (though consistent within $1\text{--}2\sigma$ ) and that systematic effects, such as the omission of unclassified sources and potential oversimplification of detection efficiency in the visibility window calculation, may contribute to this discrepancy. They explicitly state that the physical interpretation of the CC SN rates in the context of SFRD history is detailed in a companion paper (Vassallo et al.), and the construction of DTDs for SNe Ia is left for future work.

The Big Discovery: Counting the Cosmic Fireworks

The "One-Shot" Challenge: Guessing with a Single Snapshot

Two Lists, One Story

What They Didn't Find (and What They Ruled Out)

The Bottom Line

More like this