Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Search for dark matter annihilation in the center of the Earth with 8 years of IceCube data

T0 review · 3 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read After eight years of IceCube data, the projected sensitivity for dark-matter neutrinos from Earth's center improves by roughly 3.8.

desk verdict Honest proceedings status report: the 3.8x sensitivity gain is a projected ceiling from MC with an optimized BDT cut, not a measured result, but the paper is transparent about that. read the letter →

arxiv 1908.07255 v1 pith:NMMF7PIX submitted 2019-08-20 astro-ph.HE hep-ph

classification astro-ph.HEhep-ph
keywords darkmatterannihilationEarthcaptureIceCubeWIMPneutrinotelescopeboosteddecisiontreesensitivityspin-independentscattering
topics Dark Matter
open problems Dark Matter
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

Dark matter particles that scatter and lose energy inside Earth can accumulate at its center; annihilations there would produce a steady neutrino flux emerging from the planet's interior. This paper develops a search for that flux in eight years of IceCube data using a boosted decision tree, a machine-learning classifier, to separate signal-like neutrino events from atmospheric muons and neutrinos. The central result is a projected sensitivity, not a measured limit: for the $\chi\chi\to W^+W^-$ channel at $m_\chi=1$ TeV, the median expected 90% upper limit on the volumetric neutrino flux is about 3.8 times better than IceCube's previous one-year analysis, and the sensitivity on the annihilation rate improves on earlier IceCube and ANTARES limits. If the real data contain no signal, these sensitivities would translate into the strongest constraints yet from the Earth-capture channel on dark-matter-nucleon spin-independent scattering.

What carries the argument

The carrying object is the BDT classifier, which assigns every event a single signal-likeness score; one BDT is trained on a $m_\chi=50$ GeV $\tau^+\tau^-$ benchmark and another on $m_\chi=1$ TeV $W^+W^-$, with the high-mass model used for the reported sensitivities. After a cut on the score, a binned Poisson likelihood over reconstructed zenith angle compares data with a signal-plus-background model built from CORSIKA air-shower muons, NuGen/GENIE atmospheric neutrinos, and oscillation treatment below 100 GeV, while the signal is generated with WimpSim. The detector-level quantity that carries the result is the volumetric neutrino flux $\Gamma_{\nu\to\mu} = N_{\rm sig}^{0.9}/(t_{\rm live}V_{\rm eff})$, which is converted through simulation to the annihilation rate $\Gamma_A$, the physical variable compared with other experiments.

What would settle it

Unblind the full eight-year signal region and compare the observed post-cut event rate and zenith-angle distribution with the simulated 0.19 mHz background: any deviation beyond the quoted uncertainties would invalidate the projected median limit. Alternatively, compute the 90% upper limit directly from the observed data; if it is statistically consistent with the previous one-year IceCube limit rather than about 3.8 times more sensitive, the central claim is falsified.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that a BDT-based event selection over eight years makes the Earth-center dark-matter channel a substantially more powerful neutrino search than the previous one-year IceCube analysis. For the benchmark channel $\chi\chi\to W^+W^-$ with $m_\chi = 1$ TeV, the median 90% sensitivity on the volumetric neutrino flux exceeds the earlier result by roughly a factor of 3.8, and the corresponding sensitivity on the annihilation rate is markedly better than both IceCube's previous upper limit and the ANTARES limit in the same channel. These numbers come from a binned likelihood in reconstructed zenith angle, a frequentist upper-limit prescription, and $10^4$ background-only pseudo-experiments, after a BDT score cut near 0.30 leaves a total background rate of 0.19 mHz. The paper presents this as the first sensitivity study of an ongoing analysis and expects further gains from moving to an event-wise unbinned likelihood that also uses reconstructed energy.

Load-bearing premise

The projection assumes that the simulated atmospheric background reproduces the real detector data after the BDT cut well enough that the median expected limit from background-only pseudo-experiments is meaningful; the paper checks that agreement on only 10% of the data and shows only statistical uncertainties in the comparison plots.

Editorial extensions

If this is right

  • If the projected sensitivity becomes the observed result, IceCube's eight-year $W^+W^-$ search would surpass both its own one-year limit and ANTARES in that channel.
  • A null signal would convert the improved sensitivity into tighter upper limits on the spin-independent dark-matter-nucleon scattering cross-section, approaching the strongest constraints available from Earth-capture searches.
  • The same BDT and likelihood chain applies to other annihilation channels; the paper reports comparable selection performance for channels such as $b\bar b$, so the method can be reused across final states.
  • Replacing the binned zenith-angle likelihood by the event-wise unbinned version with reconstructed energy is expected, as the paper states, to improve sensitivity further.

Reading between the lines

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

  • The quoted 3.8-fold gain is benchmarked at $m_\chi=1$ TeV; a plausible extension, not stated in the paper, is that the gain varies with mass and may be smaller at low masses where the softer neutrino spectrum makes DeepCore veto performance the limiting factor.
  • Because the background estimate is validated on only 10% of the data, a natural robustness check is to compare the post-cut rate (0.19 mHz) and zenith distribution year by year; if the rate drifts, the median-limit projection degrades in a mass-dependent way.
  • The likelihood structure transfers directly to any fixed-direction source, so applying the same eight-year data set and BDT chain to the Sun would likely produce a comparable sensitivity gain for solar WIMP capture.
  • Adding reconstructed energy to the likelihood is a natural test of the method: the energy spectrum of secondary neutrinos differs sharply between $W^+W^-$ and $b\bar b$ final states, so an energy-aware version of this analysis should separate channels more cleanly than the zenith-only version.
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.

Referee Report

3 major / 5 minor

Summary. The paper reports an updated sensitivity study for a search for dark matter annihilation in the center of the Earth using 8 years of IceCube data (2011-2018). The analysis employs a BDT-based event selection trained on two benchmark signals (mχ=50 GeV, τ+τ−; mχ=1 TeV, W+W−), a binned likelihood in the reconstructed zenith angle, and the Feldman-Cousins method to derive a 90% sensitivity defined as the median upper limit from 10^4 background-only pseudo-experiments. The central result is a projected volumetric neutrino-flux sensitivity that exceeds the previous IceCube result by a factor of ~3.8 at mχ=1 TeV in the W+W− channel, together with improved sensitivity on the annihilation rate compared to previous IceCube and ANTARES limits. The paper also outlines a future extension to an event-wise unbinned likelihood including reconstructed energy.

Significance. If the projected improvement is robust, this analysis would place IceCube in a competitive position for spin-independent dark matter-nucleon scattering constraints from Earth-capture signals, a complementary probe to direct-detection and solar-capture searches. The work uses standard, publicly documented simulation tools (CORSIKA, NuGen/GENIE, WimpSim), a well-established statistical framework (binned likelihood, Feldman-Cousins), and a large background-only pseudo-experiment ensemble, which provides a reproducible basis for the sensitivity estimate. These strengths are, however, partly offset by the issues identified in the major comments, so the quantitative significance of the claimed gain is currently not fully established.

major comments (3)
  1. [Section 5, central paragraph (after Fig. 4)] The BDT score cut is set to 0.30 by a preliminary scan that optimizes the sensitivity on the same Monte Carlo used to compute the reported median limit. This in-sample optimization procedure introduces a selection bias that makes the quoted 3.8x improvement over the previous IceCube result a best-case ceiling rather than an unbiased expectation. The authors should either demonstrate that the sensitivity is stable under reasonable variations of the cut, use a validation sample that was not used for cut selection, or quantify the trial factor associated with the scan. Without this, the central claim is not fully robust.
  2. [Section 4 and Figs. 2 and 3 captions] The background estimate after the final BDT cut (0.19 mHz, Ntot=43032 events) and the zenith-angle background PDF B(bini) entering Eq. (5.1) are derived entirely from Monte Carlo (CORSIKA atmospheric muons, NuGen/GENIE neutrinos, and oscillation treatment), with the captions of Figs. 2 and 3 explicitly stating that only statistical uncertainties are included. Data/MC agreement in the signal region is validated with only a 10% data subset, which is then disregarded. Unmodeled systematics in the background rate or in the PDF shape, especially near the Earth-center direction (cos θ≈−1) where the signal peaks, directly propagate into the likelihood and would degrade the realized limits. The paper should propagate at least conservative systematic uncertainties on the normalization and shape of B(bini), or clearly label the reported sensitivity as a statistical-only upper bound.
  3. [Section 5, Fig. 5 and Sec. 1] The claimed improvement factor of ~3.8 compares a median expected sensitivity (from pseudo-experiments) with 'the previous IceCube sensitivity [2]'. If reference [2] actually reports an observed upper limit, as suggested by the phrase 'published upper limits on the spin-independent DM-nucleon scattering cross-section' in Sec. 1, then the comparison is not on equal footing: a median sensitivity should be compared with the previous analysis's median expected sensitivity, not with its observed limit. Please clarify the quantity taken from [2] and, if needed, recompute the improvement factor so that the comparison is apples-to-apples.
minor comments (5)
  1. [Section 2, DeepCore description] The text states that DeepCore is installed at 'a depth of 1750 cm'; the intended unit is presumably meters (1750 m), consistent with the surrounding depths quoted in meters. Please correct this typo.
  2. [Section 4, event selection paragraph] The sentence 'During the development of the event selection is was verified...' contains a typo ('is was' should be 'it was'), and the later sentence 'An initial series of cuts is applied on to reduce...' is grammatically awkward ('on to' should likely be 'to'). Please proofread these passages.
  3. [Section 5, Eq. (5.1)] The notation of the product index 'binmax ∏ bini=binmin' is nonstandard and slightly confusing; typically one writes ∏_{bini=binmin}^{binmax}. The authors may want to clarify that 'bini' is a bin label, not a running index.
  4. [Section 6, Conclusion] The conclusion says the new analysis 'could lead to world competitive limits' on the spin-independent cross-section, but the paper presents sensitivities, not observed limits. The wording should be adjusted to say that the analysis could produce world-competitive sensitivities or limits once applied to the full data.
  5. [General] The paper would benefit from an explicit statement that all quoted results are median expected sensitivities and not observed limits, and from specifying how the 10% data validation subset is handled (the statement that it is 'disregarded for further analysis' is helpful but could be made more precise, e.g., that it is excluded from both the sensitivity estimate and any future limit calculation).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the sensitivity projection is MC-based but is not definitionally tied to its inputs; the BDT optimization is a statistical caveat, not a circular step.

full rationale

The paper reports an expected sensitivity, defined as the median 90% upper limit from 10^4 background-only pseudo-experiments. The signal and background zenith-angle PDFs entering Eq. 5.2 are produced by independent external simulations (WimpSim for signal; CORSIKA, NuGen/GENIE for background), not by fitting the reported limit itself. The ~3.8x improvement over the previous IceCube sensitivity [2] is a comparison between two separately computed sensitivity curves, and the comparison with ANTARES [12] is an external benchmark; neither reduces to a fitted parameter being renamed as a prediction. The only in-sample optimization is the BDT score cut, chosen by scanning the sensitivity on the same MC; this can make the quoted sensitivity optimistic if the MC background is inaccurate, and the paper itself notes that only statistical uncertainties are shown in Figs. 2-3 and that only ~10% of data is used for data/MC validation in the signal region. These are legitimate statistical/systematic caveats, not circularity, because no equation in the paper equates the output to an input by construction. Self-citation to [2] is limited to methodological similarity and baseline comparison and is not load-bearing.

Assumptions & free parameters 1 free parameters · 5 assumptions · 0 invented entities

The sensitivity projection rests on a standard halo model, the Griest-Seckel capture-annihilation relation, WimpSim signal spectra, and a simulation-based background model validated on 10% of the data. There are no newly postulated particles or forces; the only tuned parameter is the BDT score cut, chosen on the metric being reported.

free parameters (1)
  • BDT score cut value = 0.30
    Selected to maximize the projected sensitivity (Sec. 5). Because it is tuned on the same simulation that produces the quoted 3.8x improvement, the gain could be optimistic.
assumptions (5)
  • domain assumption Standard Halo Model: truncated Maxwellian velocity distribution with dispersion 270 km/s, escape velocity 544 km/s, and local dark matter density 0.3 GeV/cm3
    Used in Sec. 3 (Fig. 1, taken from [7]) to compute the capture rate and hence the expected annihilation flux; the cited literature [6] notes the local density can vary from about 0.2 to 0.5 GeV/cm3.
  • domain assumption Capture-annihilation equilibrium relation Gamma_A = (C/2) tanh^2(t_Earth/tau), Eq. (3.1)
    Taken from Griest and Seckel [4]. If equilibrium is not reached for low cross-sections, the sensitivity on the annihilation rate does not translate linearly into a limit on the scattering cross-section.
  • domain assumption MC background simulations (CORSIKA, NuGen, GENIE) with a 10% data validation subset correctly model the data after the BDT cut
    Section 4; Figs. 2 and 3 show data/MC agreement with statistical uncertainties only, so unmodeled systematics could shift the projected sensitivity.
  • domain assumption WimpSim neutrino spectra correctly describe the annihilation channels
    Used both to build the signal PDFs S(bini) and to convert the flux to annihilation rates (Secs. 4 and 5). The projected limit is therefore dependent on the modeled annihilation spectra.
  • standard math Feldman-Cousins 90% upper limit construction with a binned Poisson likelihood
    Eqs. (5.1) and (5.2) in Sec. 5; standard frequentist machinery applied to pseudo-experiments, no novel statistical claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Search for dark matter annihilation in the center of the Earth with 8 years of IceCube data." pith.science (2026). https://pith.science/paper/NMMF7PIX

@misc{pith2026190807255,
  author       = {Pith},
  title        = {Pith review of: Search for dark matter annihilation in the center of the Earth with 8 years of IceCube data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NMMF7PIX}},
  note         = {Machine review of arXiv:1908.07255}
}
read the original abstract

Dark matter particles in the galactic halo can scatter off particles in celestial bodies such as stars or planets, lose energy and become gravitationally trapped. In this process, an accumulation of dark matter in the center of celestial bodies is expected, for example, at the center of the Earth. If dark matter self-annihilates into Standard Model particles, the end products of these annihilations include neutrinos. The IceCube Neutrino Observatory at the geographic South Pole can detect the resulting flux of neutrinos originating from dark matter annihilation in the center of the Earth. A search for this signal is on-going using 8 years of IceCube data and probing different annihilation channels. Here the sensitivities are presented for this new analysis, showing significant improvements with respect to the previous analyses from IceCube and other experiments.

Figures

Figures reproduced from arXiv: 1908.07255 by the authors.

Figure 1
Figure 1. Capture rate of DM particles at Earth assuming σSI = 10−42 cm2 . From [7]. The neutrino flux arising from DM annihilation in the center of the Earth is given by dΦ dEν = ΓA 4πR 2 ⊕ dNν dEν , (3.2) where R⊕ is the Earth radius and dNν /dEν denotes the energy spectrum of secondary neutrinos produced in these annihilations, which depends on the WIMP mass and annihilation channel. 4. Data and simulations In this work 8 … view at source ↗
Figure 2
Figure 2. BDT Scores obtained with a BDT trained on mχ = 50 GeV and annihilation channel χχ → τ +τ −. The violet lines indicate the two benchmark signal distributions. Distributions of atmospheric muons and neutrinos are shown in pink and green, respectively. Total MC background and experimental data are shown as a grey band and a dotted black line, respectively. Only statistical uncertainties are included [PITH_FULL_IMAGE:f… view at source ↗
Figure 3
Figure 3. BDT Scores obtained with a BDT trained on mχ = 1 TeV and annihilation channel χχ →W+W−. The violet lines indicate the two benchmark signal distributions. Distributions of atmospheric muons and neutrinos are shown in pink and green, respectively. Total MC background and experimental data are shown as a grey band and a dotted black line, respectively. Only statistical uncertainties are included. 5 [PITH_FULL_IMAGE:fi… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Background PDF (orange) and signal PDF (blue) for channel χχ → W+W−, mχ = 1 TeV. Final high mass BDT cut value is 0.30 A work to improve the previously presented analysis method is currently being developed. Firstly, the likelihood can be calculated on an event-by-even…
Figure 5
Figure 5. Figure 5: Estimated sensitivity (blue solid line) on the conversion rate for the annihilation channel χχ → W+W− with a BDT cut of 0.30, compared to the previous IceCube result [2] (orange) [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Sensitivity on the annihilation rate for the annihilation channel χχ →W+W−, compared with the upper limits from IceCube [2] (blue) and ANTARES [12] (orange). 7 [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Complementary Planetary Spectroscopy Probes of Dark Matter

    hep-ph 2025-08 conditional novelty 6.0 of 10

    Dark matter annihilation energy deposited in planetary atmospheres and interiors, compared against existing UV airglow and heat flow measurements, yields new sub-GeV scattering constraints and long-lived mediator reach.

Reference graph

Works this paper leans on

12 extracted references · 11 canonical work pages · cited by 1 Pith paper

  1. [2]

    IceCube Collaboration, M. G. Aartsen et al., EPJ C 77 (2017) 82

  2. [1]

    Bertone, D

    G. Bertone, D. Hooper, and J. Silk, Phys.Rept. 405 (2005) 279–390

  3. [3]

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

  4. [4]

    Griest and D

    K. Griest and D. Seckel, Nucl. Phys. B 283 (1987) 681

  5. [5]

    Jungman, M

    G. Jungman, M. Kamionkowski, and K. Griest, Phys. Rept. 267 (1996) 195–373

  6. [6]

    J. I. Read, J. Phys. G 41 (2014)

  7. [7]

    Sivertsson and J

    S. Sivertsson and J. Edsjö, Phys. Rev. D 85 (2012)

  8. [8]

    D. Heck, J. Knapp, J. Capdevielle, G. Schatz, and T. Thouw, FZKA 6019 (1998)

Show all 12 references
  1. [9]

    IceCube Collaboration, M. G. Aartsen et al., Phys. Rev. Lett. 120 (2018) 071801

  2. [10]

    Blennow, J

    M. Blennow, J. Edsjö, and T. Ohlsson, JCAP WimpSim Neutrino Monte Carlo, http://wimpsim.astroparticle.se/ (2008) 021

  3. [11]

    G. J. Feldman and R. D. Cousins, Phys. Rev. D 57 (Apr, 1998) 3873–3889

  4. [12]

    Dark Univ

    ANTARES Collaboration, Phys. Dark Univ. 16 (2017) 41–48. 8

Pith tools

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