Searching for Time-Dependent Neutrino Emission from Blazars with IceCube

Chad Finley (for the IceCube Collaboration); Erin O'Sullivan

Searching for Time-Dependent Neutrino Emission from Blazars with IceCube

T0 review · 0 major / 7 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper finds no statistically significant population of blazar neutrino flares in five years of IceCube muon-neutrino data, with final p=0.244 excluding TXS 0506+056 and p=0.114 including it.

desk verdict A careful, calibrated null result from the first systematic time-dependent neutrino flare search over the northern 3LAC catalog; method is borrowed but the survey and its ensemble calibration are new. read the letter →

arxiv 1908.05526 v1 pith:DEA67KZ3 submitted 2019-08-15 astro-ph.HE hep-ex

Erin O'Sullivan , Chad Finley (for the IceCube Collaboration) This is my paper · ORCID

classification astro-ph.HEhep-ex

keywords blazarsneutrinoflaresmuonneutrinosIceCubetime-dependentanalysisunbinnedmaximumlikelihoodbinomialtest3LACcatalog

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

After the 2017 coincidence between a high-energy neutrino and the flaring blazar TXS 0506+056, the question is whether other blazars also produce neutrino flares. This paper applies the same time-dependent flare search used to find the 2014-15 neutrino flare from TXS 0506+056 to every one of the 1023 northern-sky blazar and AGN directions in the Fermi 3LAC catalog, fitting the most significant neutrino flare in each direction without reference to gamma-ray light curves. It then asks whether the distribution of the 1023 best-fit p-values contains more extremely small values than background would produce, using a binomial test designed to catch a small population of neutrino emitters. The test finds no such excess: after removing correlated directions, the final significance is $p=0.244$ when TXS 0506+056 is excluded and $p=0.114$ when it is included, both consistent with atmospheric background. If correct, the result means that five years of IceCube muon-neutrino data show no evidence that blazars as a class are episodic neutrino emitters beyond TXS 0506+056.

What carries the argument

The engine of the search is the unbinned maximum-likelihood flare fit of equations (2.1)-(2.4): a likelihood in which a signal component with spatial, energy, and Gaussian time terms (equation 2.2) is mixed with an atmospheric-background term (equation 2.3), and the test statistic $D$ penalizes short time windows through the livetime-to-window ratio $\mathrm{TL}/\hat{T}_\sigma$ to account for the increased trial factor. Its output is a best-fit number of signal events, spectral index, flare time, and flare width, which is converted to a local p-value against scrambled data. The second piece of machinery is the binomial test of equation (2.5), which scans the 100 smallest ordered p-values and asks whether $k$ or more blazars have p-values at or below $p_k$. The third piece is a correlation-removal prescription: neutrino events are ranked by signal-to-background ratio $S_i/B_i$, two search directions are called correlated if they share any of the top $3\hat{n}_s$ neutrinos, and $k$ is reduced by the number of redundant directions before the p-value is recomputed. Finally, the entire decorrelated-binomial procedure is run on scrambled data to produce the null distribution against which the data's decorrelated p-value is judged.

What would settle it

Run the identical pipeline on simulated data in which ten of the 1023 blazar directions each inject a five-event Gaussian neutrino flare: if the decorrelated binomial p-value is not systematically smaller than the scrambled-background distribution, then the analysis cannot detect the population it appears to exclude, and the reported null result would not constrain small populations.

Watch

Extended reading notes

Core claim

On the paper's own terms: for each of the 1023 search directions, the best-fit neutrino flare is found by maximizing an unbinned likelihood in which the signal is a spatially Gaussian, energy-power-law, time-Gaussian burst and the background is atmospheric muon neutrinos; the test statistic penalizes short time windows for their larger trial factor. The p-value of each direction's best fit is calibrated against scrambled-data ensembles. The smallest binomial p-value over the 100 lowest p-values occurs at $k=11$, with local p-value $3.27\times10^{-3}$. Two correlation clusters are identified in those 11 directions (three directions around B2 1126+37 and two around GB6 J0929+5013) and removed by lowering $k$ to 8, giving a decorrelated binomial p-value of $2.09\times10^{-2}$ including TXS 0506+056 and $5.33\times10^{-2}$ without it. When these decorrelated values are compared with the same quantity from scrambled data, the final p-values are $0.114$ and $0.244$, respectively, so the data are consistent with background. The same test applied separately to FSRQ and BLLac sub-catalogs gives final p-values of $0.50$ and $0.60$.

Load-bearing premise

Everything rests on scrambled events faithfully reproducing the atmospheric-neutrino background: if scrambling distorts detector acceptance, energy response, or the spatial-time correlation structure, every reported p-value is biased.

Editorial extensions

If this is right

At p=0.244 excluding TXS 0506+056, the observed set of per-direction flare p-values is consistent with atmospheric-neutrino background alone.
The eight uncorrelated directions that survive decorrelation in the top 11 do not constitute a significant population; the analysis places no limit on individual sources but bounds the collective excess.
FSRQ-only and BLLac-only sub-catalogs each return final p-values consistent with background (0.50 and 0.60).
The three non-blazar objects among the top 11 are a small a posteriori fraction, and the paper treats them as motivation for follow-up rather than evidence.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A natural extension the paper leaves implicit: the same catalog scan with time windows fixed by gamma-ray flares could be more sensitive to blazars that emit neutrinos only while gamma-bright, since the present search deliberately fits time without any electromagnetic trigger.
Because the final p-value comes from comparing the data's decorrelated binomial p-value with scrambled skies, the claim is about the population of detectable flares: a population whose flares are too weak to enter the top 100 p-value list would be invisible to this test.
One testable improvement would be to run the same binomial test on an independent half of the data to check that the reported p=0.244 reproduces.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

This is a well-behaved null result from the first systematic search for time-dependent neutrino flares across the northern-sky 3LAC catalog. The likelihood machinery is taken from earlier IceCube papers, so methodological novelty is low, but the survey itself is new: they fit the most significant flare in each of 1023 directions, apply a binomial test to the p-values, decorrelate nearby directions that share events, and calibrate the final p-value against scrambled data. The headline number is p=0.244 excluding TXS 0506+056 (0.114 including), consistent with background. That is a defensible statement that the 5-year muon-neutrino sample contains no significant population of blazar neutrino flares.

What the paper does well: the calibration is careful. Rather than quoting the raw binomial p-value, they compare it to the distribution of the same decorrelated statistic on scrambled data, which absorbs residual non-uniformities from the decorrelation procedure. They report with and without the motivating source, and they separately test FSRQ and BL Lac sub-catalogs. The per-object fit table is useful.

Soft spots: the decorrelation recipe (top ns*3 neutrinos, reduce k by the number of correlated directions) is ad hoc, though applying it identically to data and scrambles makes the final p-value valid under the null. The scan over k=1..100 is a choice, not derived from a prior on population size. More substantively, no detection sensitivity is quoted, so the null result is not as sharp as it could be—readers can't tell what fraction of a putative blazar population would have been excluded. The analysis code and data are not public, which limits reproducibility, though the method is documented in prior papers. The authors themselves flag the non-blazar counterpart observation as post hoc, which is the right call.

None of this undermines the central claim. The scrambled-data null is the main vulnerability, but no specific failure of that assumption surfaces here, and the final calibration is a standard Monte Carlo move. This is a valid contribution for multimessenger astrophysics, especially as a companion to the TXS 0506+056 results. It deserves a serious referee; the main requests would be sensitivity numbers and more detail on the decorrelation. I'd cite it if I were working on blazar neutrino searches.

Recommendation: send to peer review.

Referee Report

0 major / 7 minor

Summary. This paper presents a search for time-dependent neutrino emission from blazars in the Fermi 3LAC catalog, using 5 years of IceCube muon-neutrino data. For each of 1023 catalog directions, the best-fit neutrino flare is found with an unbinned maximum-likelihood method, and a p-value is assigned by comparing the test statistic to scrambled data. A binomial test is then applied to the set of p-values to look for an excess of low p-values beyond the background expectation. Correlated search directions are removed with a heuristic prescription, and the final binomial p-value is calibrated by repeating the entire procedure on scrambled data. The resulting final p-values are 0.114 with TXS 0506+056 included and 0.244 with it excluded, consistent with the background-only hypothesis. Sub-catalog tests for FSRQ and BL Lac objects also show no significant excess. The most significant fitted flare is TXS 0506+056, with a local p-value of 3.47e-5.

Significance. If the result is correct, it places a meaningful constraint on the fraction of 3LAC blazars with flaring neutrino emission: the data show no evidence for a population of neutrino-bright blazars beyond the known TXS 0506+056. The analysis is a clear, well-calibrated null result. Its main strength is the use of scrambled data not only to assign individual p-values but also to calibrate the final binomial p-value, which absorbs the trial factor from scanning over k and the effects of correlations. The paper is transparent in reporting results with and without TXS 0506+056 and for the two optical sub-classes. The individual fit parameters in Table 1 are useful for follow-up. The central claim is defensible and appropriately limited.

minor comments (7)

[2.3] The algorithm for determining how many AGNs in the top k need to be removed is not fully specified; please describe the exact removal procedure (for example, greedy removal or maximum independent set) so that the analysis is reproducible.
[2.3] The threshold for a 'significant neutrino contributor' is defined as the top ns×3 neutrinos by Si/Bi; the factor of 3 appears arbitrary. Please provide a justification or demonstrate that the final p-value is insensitive to this choice.
[2.2] The scan over the 100 lowest p-values is an arbitrary upper bound; please state whether the final result changes if the scan is restricted to, for example, k=1..50 or extended to k=1..200, since the calibration only absorbs the scan range actually used.
[2.1] The scrambling procedure is described only as 'randomized positions and times'; please specify which components of the data are preserved (for example, declination, energy, livetime) or cite a reference that does so, because the null distribution depends on this.
[3.1] The raw binomial p-value for k=11 before decorrelation is not stated; reporting it would make the effect of the correlation-removal step explicit and would help the reader interpret the size of the correction.
[2.1] In the sentence describing the signal PDF, 'spacial' should be 'spatial'.
[Figure 3] The axes in Figure 3 are not labeled in the manuscript; adding x- and y-axis labels would aid interpretation.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the final p-value is produced by Monte Carlo calibration against scrambled data and is not forced by construction.

full rationale

The derivation chain is self-contained against an internal scrambled-data benchmark. The per-direction p-values (Sec. 2.1) are outputs of an unbinned likelihood maximization, converted to p-values by comparing the test statistic D with the distribution from scrambled data; they are not fitted parameters that are subsequently renamed as predictions. The binomial test (Sec. 2.2) is a standard combinatorial calculation on those p-values, and the correlation-removal prescription (Sec. 2.3) is applied identically to data and scrambles before the final comparison. The final p = 0.244 (0.114 with TXS) is therefore a calibration of the observed binomial statistic against its null distribution, not a quantity forced by construction. The only load-bearing assumption is that scrambling represents the background null, but that is a statistical assumption, not circularity: no claim in the paper defines a quantity in terms of the result it is supposed to test, and no fitted input is called a prediction. Self-citations to the time-dependent likelihood formalism ([7],[8]) and previous TXS analysis ([1],[2]) supply the method and motivation, but the present ensemble test does not reduce to those citations; it evaluates an independent question.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the standard IceCube analysis framework: the Gaussian time-window signal model, the atmospheric background spectrum, and the use of scrambled event sets as the null distribution. The only hand-chosen analysis settings are the decorrelation threshold (ns*3) and the binomial scan range (k=1..100). No new entities are introduced.

free parameters (2)

Correlation threshold factor (ns*3) = 3
Used in Section 2.3 to define which neutrinos are significant contributors when identifying correlated blazar directions. Chosen by hand; larger or smaller values would change which directions are removed.
Binomial scan range k = 1 to 100 = 100
Section 2.2 scans the 100 lowest p-values to search for an excess. The final p-value is calibrated on scrambled data, which partially absorbs this choice, but it remains a hand-set analysis parameter.

assumptions (4)

domain assumption Neutrino flares have a Gaussian time profile with unknown centre and width (Eq. 2.2).
If real blazar flares have sharply different temporal shapes or durations outside the searched range, the likelihood test would be suboptimal and could miss them.
domain assumption Scrambled event sets (randomized positions and times) faithfully reproduce the null distribution of the test statistic (Sections 2.1 and 2.3).
The entire p-value calibration rests on the assumption that scrambling preserves detector acceptance, energy response, and correlation structure of background.
domain assumption Atmospheric neutrino energy spectrum P_atm(E) accurately describes the background (Eq. 2.3).
Errors in the background spectrum model would propagate into the likelihood ratio and p-values.
domain assumption The test statistic D in Eq. 2.4, with the T_L / T_sigma penalty, is a valid trials-corrected statistic for scanning flare durations.
This analytic penalty, together with the scrambled-data calibration, must jointally capture the true trials factor for the scanned time windows.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Searching for Time-Dependent Neutrino Emission from Blazars with IceCube." pith.science (2026). https://pith.science/paper/DEA67KZ3

@misc{pith2026190805526,
  author       = {Pith},
  title        = {Pith review of: Searching for Time-Dependent Neutrino Emission from Blazars with IceCube},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DEA67KZ3}},
  note         = {Machine review of arXiv:1908.05526}
}

read the original abstract

In 2017, IceCube detected a high energy neutrino in coincidence with the blazar TXS 0506+056. In a follow up analysis of archival data, evidence for previous neutrino emission from the blazar during 2014-15 was found. In this analysis, we investigate whether other blazars might have had episodes of neutrino emission. We use a 5-year sample of muon-neutrino data and the same timedependent algorithm that was used for the archival analysis of TXS 0506+056, applying it to all of the objects in the northern sky listed in the Fermi 3LAC catalog. We fit for the most significant neutrino flare from each direction, without assuming any correlation with electromagnetic light curves. To search for an excess of significant flares which could indicate a small population of neutrino emitters, we apply a binomial test. This binomial test does not find a significant excess of flares relative to the results expected for background coincidences from atmospheric neutrinos. The final significance of the analysis is p=0.244 when TXS 0506+056 is excluded, consistent with the background-only hypothesis. We report the parameters of the individual fits for the most significant objects which contributed to the final result.

Figures

Figures reproduced from arXiv: 1908.05526 by the authors.

**Figure 1.** A flowchart showing the steps of this analysis. The signal PDF can be written as Si = 1 2πσ2 i e − r 2 i 2σ 2 i ×P(Ei |γ)× 1 √ 2πTσ e − (Ti−TO) 2 2T 2 σ . (2.2) The first term represents the spatial component of the likelihood, where reconstructed neutrino event i has an angular resolution σi and an angular separation with the search direction ri . The second term, P(Ei |γ), gives the probability distribution for t… view at source ↗

**Figure 2.** shows the distribution of p-values found from each direction in the catalog. Performing the binomial test as described in Section 2.2, we found the smallest binomial test p-value at k=11. The p-values for the top 11 3LAC directions are shown in red in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** The distribution of binomial test p-values, with correlations removed, from analysis of scrambled data (grey), compared with the value obtained in the data, including TXS 0506+056 (black) and without TXS 0506+056 (red). 3.2 Testing the Optical Class Sub-catalogs We separately tested two sub-catalogs of the 3LAC catalog: directions corresponding to blazars classified as FSRQ, and directions corresponding to blazars c… view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Skymap in equatorial coordinates of the search directions in the 3LAC catalog, shown in grey. In red, we show the 11 search directions that yield the lowest p-value in the binomial test (8 of which are unique). with neutrino emission. We report in these proceedings the…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

14 extracted references · 6 canonical work pages

[1]

write newline

" write newline "" before.all 'output.state := FUNCTION blank.sep after.quote 'output.state := FUNCTION fin.entry output.state after.quoted.block = 'skip 'add.period if write newline FUNCTION new.block output.state before.all = 'skip output.state after.quote = after.quoted.block 'output.state := after.block 'output.state := if if FUNCTION new.sentence out...
[2]

IceCube, and others Collaboration, M. G. Aartsen et al., Science 361 (2018) eaat1378

work page 2018
[3]

IceCube Collaboration, M. G. Aartsen et al., Science 361 (2018) 147--151

2018
[4]

IceCube Collaboration, M. G. Aartsen et al., JINST 12 (2017) P03012

2017
[5]

Williams, PoS(ICRC2019)16 (these proceedings)

IceCube Collaboration, D. Williams, PoS(ICRC2019)16 (these proceedings)

work page
[6]

Ackermann et al., Astrophys

Fermi-LAT Collaboration, M. Ackermann et al., Astrophys. J. 810 (2015) 14

work page 2015
[7]

IceCube Collaboration, M. G. Aartsen et al., Eur. Phys. J. C79 (2019) 234

work page 2019
[8]

Braun , J

J. Braun , J. Dumm , F. De Palma , C. Finley , A. Karle , and T. Montaruli , Astropart. Phys. 29 (2008) 299--305

work page 2008

Show all 14 references

[9]

Braun, M

J. Braun, M. Baker, J. Dumm, C. Finley, A. Karle, and T. Montaruli, Astropart. Phys. 33 (2010) 175--181

2010
[10]

IceCube Collaboration, PoS(ICRC2015)1099 (2016)

2016
[11]

IceCube-Gen2 Collaboration, PoS(ICRC2017)1052 (2018)

2018
[12]

IceCube Collaboration, PoS(ICRC2019)xyz (these proceedings)
[13]

IceCube Collaboration, M. G. Aartsen et al. , JINST 12 (2017) P03012

2017
[14]

Waxman and J

E. Waxman and J. N. Bahcall, Phys. Rev. D59 (1999) 023002

1999

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [1]

write newline

" write newline "" before.all 'output.state := FUNCTION blank.sep after.quote 'output.state := FUNCTION fin.entry output.state after.quoted.block = 'skip 'add.period if write newline FUNCTION new.block output.state before.all = 'skip output.state after.quote = after.quoted.block 'output.state := after.block 'output.state := if if FUNCTION new.sentence out...

[2] [2]

IceCube, and others Collaboration, M. G. Aartsen et al., Science 361 (2018) eaat1378

work page 2018

[3] [3]

IceCube Collaboration, M. G. Aartsen et al., Science 361 (2018) 147--151

2018

[4] [4]

IceCube Collaboration, M. G. Aartsen et al., JINST 12 (2017) P03012

2017

[5] [5]

Williams, PoS(ICRC2019)16 (these proceedings)

IceCube Collaboration, D. Williams, PoS(ICRC2019)16 (these proceedings)

work page

[6] [6]

Ackermann et al., Astrophys

Fermi-LAT Collaboration, M. Ackermann et al., Astrophys. J. 810 (2015) 14

work page 2015

[7] [7]

IceCube Collaboration, M. G. Aartsen et al., Eur. Phys. J. C79 (2019) 234

work page 2019

[8] [8]

Braun , J

J. Braun , J. Dumm , F. De Palma , C. Finley , A. Karle , and T. Montaruli , Astropart. Phys. 29 (2008) 299--305

work page 2008