Supplementary Materials for

Observation of high-energy neutrinos from the Galactic plane

IceCube Collaboration

Corresponding authors: analysis@icecube.wisc.edu; F. Halzen, francis.halzen@icecube.wisc.edu

Science 380, 1338 (2023)

DOI: 10.1126/science.adc9818

The PDF file includes:

Materials and Methods

Supplementary Text
Figs. S1 to S12
Tables S1 to S5
IceCube Collaboration Authors
References
 Supplementary materials for:
Observation of high-energy neutrinos from the Galactic Plane

IceCube Collaboration∗ :

R. Abbasi17 , M. Ackermann61 , J. Adams18 , J. A. Aguilar12 , M. Ahlers22 , M. Ahrens51 , J. M. Alameddine23 ,

A. A. Alves Jr.31 , N. M. Amin43 , K. Andeen41 , T. Anderson58 , G. Anton26 , C. Argüelles14 ,

Y. Ashida39 , S. Athanasiadou61 S. Axani15 , X. Bai47 , A. Balagopal V.39 , S. W. Barwick30 ,

V. Basu39 , S. Baur12 , R. Bay8 , J. J. Beatty20, 21 , K.-H. Becker60 , J. Becker Tjus11 , J. Beise59 ,
C. Bellenghi27 , S. Benda39 , S. BenZvi49 , D. Berley19 , E. Bernardini61, 62 , D. Z. Besson34 ,

G. Binder8, 9 , D. Bindig60 , E. Blaufuss19 , S. Blot61 , M. Boddenberg1 , F. Bontempo31 , J. Y. Book14 ,

J. Borowka1 , S. Böser40 , O. Botner59 , J. Böttcher1 , E. Bourbeau22 , F. Bradascio61 , J. Braun39 ,
B. Brinson6 , S. Bron28 , J. Brostean-Kaiser61 , R. T. Burley2 , R. S. Busse42 , M. A. Campana46 ,

E. G. Carnie-Bronca2 , C. Chen6 , Z. Chen52 , D. Chirkin39 , K. Choi53 , B. A. Clark24 , K. Clark33 ,

L. Classen42 , A. Coleman43 , G. H. Collin15 , A. Connolly20, 21 , J. M. Conrad15 , P. Coppin13 ,
P. Correa13 , D. F. Cowen57, 58 , R. Cross49 , C. Dappen1 , P. Dave6 , C. De Clercq13 , J. J. DeLaunay56 ,

D. Delgado López14 , H. Dembinski43 , K. Deoskar51 , A. Desai39 , P. Desiati39 , K. D. de Vries13 ,

G. de Wasseige36 , T. DeYoung24 , A. Diaz15 , J. C. Díaz-Vélez39 , M. Dittmer42 , H. Dujmovic31 ,
M. Dunkman58 , M. A. DuVernois39 , T. Ehrhardt40 , P. Eller27 , R. Engel31, 32 , H. Erpenbeck1 ,

J. Evans19 , P. A. Evenson43 , K. L. Fan19 , A. R. Fazely7 , A. Fedynitch55 , N. Feigl10 , S. Fiedlschuster26 ,

A. T. Fienberg58 , C. Finley51 , L. Fischer61 , D. Fox57 , A. Franckowiak11, 61 , E. Friedman19 ,
A. Fritz40 , P. Fürst1 , T. K. Gaisser43 , J. Gallagher38 , E. Ganster1 , A. Garcia14 , S. Garrappa61 ,

L. Gerhardt9 , A. Ghadimi56 , C. Glaser59 , T. Glauch27 , T. Glüsenkamp26 , N. Goehlke32 , A. Goldschmidt9 ,

J. G. Gonzalez43 , S. Goswami56 , D. Grant24 , T. Grégoire58 , S. Griswold49 , C. Günther1 , P. Gutjahr23 ,
C. Haack27 , A. Hallgren59 , R. Halliday24 , L. Halve1 , F. Halzen39 , M. Ha Minh27 , K. Hanson39 ,

J. Hardin39 , A. A. Harnisch24 , A. Haungs31 , K. Helbing60 , F. Henningsen27 , E. C. Hettinger24 ,

S1
S. Hickford60 , J. Hignight25 , C. Hill16 , G. C. Hill2 , K. D. Hoffman19 , K. Hoshina39, 63 , W. Hou31 ,
F. Huang58 , M. Huber27 , T. Huber31 , K. Hultqvist51 , M. Hünnefeld23 , R. Hussain39 , K. Hymon23 ,

S. In53 , N. Iovine12 , A. Ishihara16 , M. Jansson51 , G. S. Japaridze5 , M. Jeong53 , M. Jin14 ,

B. J. P. Jones4 , D. Kang31 , W. Kang53 , X. Kang46 , A. Kappes42 , D. Kappesser40 , L. Kardum23 ,
T. Karg61 , M. Karl27 , A. Karle39 , U. Katz26 , M. Kauer39 , M. Kellermann1 , J. L. Kelley39 ,

A. Kheirandish58 , K. Kin16 , J. Kiryluk52 , S. R. Klein8, 9 , A. Kochocki24 , R. Koirala43 , H. Kolanoski10 ,

T. Kontrimas27 , L. Köpke40 , C. Kopper24 , S. Kopper56 , D. J. Koskinen22 , P. Koundal31 , M. Kovacevich46 ,
M. Kowalski10, 61 , T. Kozynets22 , E. Krupczak24 , E. Kun11 , N. Kurahashi46 , N. Lad61 , C. La-

gunas Gualda61 , J. L. Lanfranchi58 , M. J. Larson19 , F. Lauber60 , J. P. Lazar14, 39 , J. W. Lee53 ,

K. Leonard39 , A. Leszczyńska43 , Y. Li58 , M. Lincetto11 , Q. R. Liu39 , M. Liubarska25 , E. Lohfink40 ,
C. J. Lozano Mariscal42 , L. Lu39 , F. Lucarelli28 , A. Ludwig24, 35 , W. Luszczak39 , Y. Lyu8, 9 ,

W. Y. Ma61 , J. Madsen39 , K. B. M. Mahn24 , Y. Makino39 , S. Mancina39 , I. C. Mariş12 , I. Martinez-

Soler14 , R. Maruyama44 , S. McCarthy39 , T. McElroy25 , F. McNally37 , J. V. Mead22 , K. Meagher39 ,
S. Mechbal61 , A. Medina21 , M. Meier16 , S. Meighen-Berger27 , Y. Merckx13 , J. Micallef24 ,

D. Mockler12 , T. Montaruli28 , R. W. Moore25 , K. Morik64 , R. Morse39 , M. Moulai15 , T. Mukherjee31 ,

R. Naab61 , R. Nagai16 , R. Nahnhauer61 , U. Naumann60 , J. Necker61 , L. V. Nguyễn24 , H. Niederhausen24 ,
M. U. Nisa24 , S. C. Nowicki24 , D. Nygren9, a , A. Obertacke Pollmann60 , M. Oehler31 , B. Oeyen29 ,

A. Olivas19 , E. O’Sullivan59 , H. Pandya43 , D. V. Pankova58 , N. Park33 , G. K. Parker4 , E. N. Paudel43 ,

L. Paul41 , C. Pérez de los Heros59 , L. Peters1 , J. Peterson39 , S. Philippen1 , S. Pieper60 , A. Pizzuto39 ,
M. Plum47 , Y. Popovych40 , A. Porcelli29 , M. Prado Rodriguez39 , B. Pries24 , G. T. Przybylski9 ,

C. Raab12 , J. Rack-Helleis40 , A. Raissi18 , M. Rameez22 , K. Rawlins3 , I. C. Rea27 , Z. Rechav39 ,

A. Rehman43 , P. Reichherzer11 , R. Reimann1 , G. Renzi12 , E. Resconi27 , S. Reusch61 , W. Rhode23 ,
M. Richman46 , B. Riedel39 , E. J. Roberts2 , S. Robertson8, 9 , G. Roellinghoff53 , M. Rongen40 ,

C. Rott50, 53 , T. Ruhe23 , D. Ryckbosch29 , D. Rysewyk Cantu24 , I. Safa14, 39 , J. Saffer32 , D. Salazar-

Gallegos24 , P. Sampathkumar31 , S. E. Sanchez Herrera24 , A. Sandrock23 , M. Santander56 , S. Sarkar25 ,

S2
S. Sarkar45 , K. Satalecka61 , M. Schaufel1 , H. Schieler31 , S. Schindler26 , T. Schmidt19 , A. Schneider39 ,
J. Schneider26 , F. G. Schröder31, 43 , L. Schumacher27 , G. Schwefer1 , S. Sclafani46 , D. Seckel43 ,

S. Seunarine48 , A. Sharma59 , S. Shefali32 , N. Shimizu16 , M. Silva39 , B. Skrzypek14 , B. Smithers4 ,

R. Snihur39 , J. Soedingrekso23 , A. Sogaard22 , D. Soldin43 , C. Spannfellner27 , G. M. Spiczak48 ,
C. Spiering61 , M. Stamatikos21 , T. Stanev43 , R. Stein61 , J. Stettner1 , T. Stezelberger9 , B. Stokstad9 ,

T. Stürwald60 , T. Stuttard22 , G. W. Sullivan19 , I. Taboada6 , S. Ter-Antonyan7 , J. Thwaites39 ,

S. Tilav43 , F. Tischbein1 , K. Tollefson24 , C. Tönnis54 , S. Toscano12 , D. Tosi39 , A. Trettin61 ,
M. Tselengidou26 , C. F. Tung6 , A. Turcati27 , R. Turcotte31 , C. F. Turley58 , J. P. Twagirayezu24 ,

B. Ty39 , M. A. Unland Elorrieta42 , N. Valtonen-Mattila59 , J. Vandenbroucke39 , N. van Eijndhoven13 ,

D. Vannerom15 , J. van Santen61 , J. Veitch-Michaelis39 , S. Verpoest29 , C. Walck51 , W. Wang39 ,
T. B. Watson4 , C. Weaver24 , P. Weigel15 , A. Weindl31 , M. J. Weiss58 , J. Weldert40 , C. Wendt39 ,

J. Werthebach23 , M. Weyrauch31 , N. Whitehorn24, 35 , C. H. Wiebusch1 , N. Willey24 , D. R. Williams56 ,

M. Wolf39 , G. Wrede26 , J. Wulff11 , X. W. Xu7 , J. P. Yanez25 , E. Yildizci39 , S. Yoshida16 , S. Yu24 ,
T. Yuan39 , Z. Zhang52 , P. Zhelnin14

1
III. Physikalisches Institut, RWTH Aachen University, D-52056 Aachen, Germany
2
Department of Physics, University of Adelaide, Adelaide, 5005, Australia
3
Department of Physics and Astronomy, University of Alaska Anchorage, Anchorage, AK
99508, USA
4
Department of Physics, University of Texas at Arlington, Arlington, TX 76019, USA
5
The Center for Theoretical Studies of Physical Systems, Clark-Atlanta University, Atlanta,
GA 30314, USA
6
School of Physics and Center for Relativistic Astrophysics, Georgia Institute of Technology,

Atlanta, GA 30332, USA

7
Department of Physics, Southern University, Baton Rouge, LA 70813, USA

S3
8
Department of Physics, University of California, Berkeley, CA 94720, USA
9
Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA
10
Institut für Physik, Humboldt-Universität zu Berlin, D-12489 Berlin, Germany
11
Fakultät für Physik & Astronomie, Ruhr-Universität Bochum, D-44780 Bochum, Germany
12
Université Libre de Bruxelles, Science Faculty CP230, B-1050 Brussels, Belgium
13
Vrije Universiteit Brussel, Dienst Elementary Particles, B-1050 Brussels, Belgium
14
Department of Physics and Laboratory for Particle Physics and Cosmology, Harvard Univer-
sity, Cambridge, MA 02138, USA
15
Department of Physics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
16
Department of Physics and The International Center for Hadron Astrophysics, Chiba Univer-
sity, Chiba 263-8522, Japan
17
Department of Physics, Loyola University Chicago, Chicago, IL 60660, USA
18
Department of Physics and Astronomy, University of Canterbury, Private Bag 4800, Christchurch,
New Zealand
19
Department of Physics, University of Maryland, College Park, MD 20742, USA
20
Department of Astronomy, Ohio State University, Columbus, OH 43210, USA
21
Department of Physics and Center for Cosmology and Astro-Particle Physics, Ohio State

University, Columbus, OH 43210, USA

22
Niels Bohr Institute, University of Copenhagen, DK-2100 Copenhagen, Denmark
23
Department of Physics, TU Dortmund University, D-44221 Dortmund, Germany
24
Department of Physics and Astronomy, Michigan State University, East Lansing, MI 48824,
USA
25
Department of Physics, University of Alberta, Edmonton, Alberta, Canada T6G 2E1
26
Erlangen Centre for Astroparticle Physics, Friedrich-Alexander-Universität Erlangen-Nürnberg,
D-91058 Erlangen, Germany

S4
27
Physik-department, Technische Universität München, D-85748 Garching, Germany
28
Département de physique nucléaire et corpusculaire, Université de Genève, CH-1211 Genève,

Switzerland
29
Department of Physics and Astronomy, University of Gent, B-9000 Gent, Belgium
30
Department of Physics and Astronomy, University of California, Irvine, CA 92697, USA
31
Karlsruhe Institute of Technology, Institute for Astroparticle Physics, D-76021 Karlsruhe,
Germany
32
Karlsruhe Institute of Technology, Institute of Experimental Particle Physics, D-76021 Karl-

sruhe, Germany
33
Department of Physics, Engineering Physics, and Astronomy, Queen’s University, Kingston,
ON K7L 3N6, Canada
34
Department of Physics and Astronomy, University of Kansas, Lawrence, KS 66045, USA
35
Department of Physics and Astronomy, University of California, Los Angeles, Los Angeles,
CA 90095, USA
36
Centre for Cosmology, Particle Physics and Phenomenology, Université catholique de Lou-
vain, Louvain-la-Neuve, Belgium
37
Department of Physics, Mercer University, Macon, GA 31207-0001, USA
38
Department of Astronomy, University of Wisconsin–Madison, Madison, WI 53706, USA
39
Department of Physics and Wisconsin IceCube Particle Astrophysics Center, University of
Wisconsin–Madison, Madison, WI 53706, USA
40
Institute of Physics, University of Mainz, Staudinger Weg 7, D-55099 Mainz, Germany
41
Department of Physics, Marquette University, Milwaukee, WI, 53201, USA
42
Institut für Kernphysik, Westfälische Wilhelms-Universität Münster, D-48149 Münster, Ger-

many
43
Bartol Research Institute and Department of Physics and Astronomy, University of Delaware,

S5
Newark, DE 19716, USA
44
Department of Physics, Yale University, New Haven, CT 06520, USA
45
Department of Physics, University of Oxford, Parks Road, Oxford OX1 3PU, UK
46
Department of Physics, Drexel University, Philadelphia, PA 19104, USA
47
Physics Department, South Dakota School of Mines and Technology, Rapid City, SD 57701,

USA
48
Department of Physics, University of Wisconsin, River Falls, WI 54022, USA
49
Department of Physics and Astronomy, University of Rochester, Rochester, NY 14627, USA
50
Department of Physics and Astronomy, University of Utah, Salt Lake City, UT 84112, USA
51
Oskar Klein Centre and Department of Physics, Stockholm University, SE-10691 Stockholm,
Sweden
52
Department of Physics and Astronomy, Stony Brook University, Stony Brook, NY 11794-
3800, USA
53
Department of Physics, Sungkyunkwan University, Suwon 16419, Korea
54
Institute of Basic Science, Sungkyunkwan University, Suwon 16419, Korea
55
Institute of Physics, Academia Sinica, Taipei, 11529, Taiwan
56
Department of Physics and Astronomy, University of Alabama, Tuscaloosa, AL 35487, USA
57
Department of Astronomy and Astrophysics, Pennsylvania State University, University Park,
PA 16802, USA
58
Department of Physics, Pennsylvania State University, University Park, PA 16802, USA
59
Department of Physics and Astronomy, Uppsala University, Box 516, S-75120 Uppsala, Swe-
den
60
Department of Physics, University of Wuppertal, D-42119 Wuppertal, Germany
61
Deutsches Elektronen-Synchrotron, D-15738 Zeuthen, Germany
62
Università di Padova, I-35131 Padova, Italy

S6
63
Earthquake Research Institute, University of Tokyo, Bunkyo, Tokyo 113-0032, Japan
64
Computer Science Faculty, TU Dortmund University, D-44221 Dortmund, Germany

∗
E-mail: analysis@icecube.wisc.edu

†
Present address: Department of Physics, University of Texas at Arlington, Arlington, TX
76019, USA

S7
Materials and Methods
Event Selection and Reconstruction

The event selection utilizes a series of convolutional neural networks (CNNs) loosely based on

the method presented previously (15). These CNNs include classification as well as regression
tasks to obtain initial reconstructions for event properties such as direction, interaction vertex,
and deposited energy. The event properties are later refined with a dedicated reconstruction

method (16). At early stages of the selection, fast and simple CNN architectures are applied to
discard the majority of background events and thus reduce the data rate for subsequent selection
steps. These CNNs require a per-event runtime of about 1 ms and they are able to reduce the

atmospheric background by about 99.92% more than the online cascade filter (18,22), while re-
taining more than half of all signal events above 500 GeV. This reduction of background events
by over three orders of magnitude allows for larger and more complex CNN architectures to be

applied in subsequent steps. The final selection step is based on gradient boosted decision trees
(BDTs) (41) that use high-level outputs of the CNNs as input. At this stage, the atmospheric
muon background is reduced by eight orders of magnitude and neutrinos dominate the sample,

thus enabling searches for neutrino sources. The per-flavor atmospheric and astrophysical neu-
trino contributions based on simulation (42) are illustrated in Figure S1. Comparisons between
experimental data distributions and simulation show reasonable agreement as illustrated in Fig-

ure S2. The analysis does not depend on an accurate modelling of the background due to the
employed data-driven search method (see below).
The resulting cascade selection retains more than 20 times as many events as the previous

IceCube analysis (12). This arises from both an overall improved efficiency and a lower energy
threshold of about 500 GeV, compared to several TeV in earlier work, enabled by improved
selection techniques. A contribution to the increase in efficiency relative to the previous selec-

S8
 106 Total Atmospheric Total Astrophysical
Atmo νe Astro νe
Number of Events
Atmo νµ Astro νµ
104 Atmo ντ Astro ντ

102

100
103 104 105 106
Eν [GeV]

Figure S1: Neutrino flavor event distribution. Number of expected atmospheric neutrino
(Atmo ν) and astrophysical neutrino (Astro ν) events for this work calculated based on simula-
tion (42). Astrophysical neutrino contributions assume an isotropic astrophysical flux (22) and
an equal flavor ratio at Earth. The contributions of atmospheric neutrinos are obtained from the
cascade-equation solver MCEq (43), the Hillas model (H3a) for cosmic rays (44, 45) and the
SIBYLL 2.3c hadronic interaction model (46). Self-Veto (24, 25) passing fraction calculated
with NuVeto (47). Shown are the contribution from each flavor (ν + ν̄) as well as the total
number of events as a function of neutrino energy (Eν ).

tion (12) is the inclusion of events with interaction vertices near the boundary or outside of the
instrumented volume as illustrated in Figure S3. These “partially contained” and “uncontained”
events are more difficult to distinguish from background events and their reconstruction is chal-

lenging due to only observing a fraction of the deposited energy. Thus, these events were re-
moved in most previous IceCube analyses. The application of deep learning-based tools allows
us to retain a large fraction of these and other challenging events, while ensuring reconstruction

quality and a reduced level of background contamination. An example for an “uncontained”

and “contained” cascade event is shown in Figure S4. About 17.5% of all events in the sample
have a reconstructed interaction vertex outside of the instrumented volume and are thus consid-

ered as uncontained events. An additional 2.5% of events are located in a region with increased
dust impurities in the ice. Despite an increased fraction of more challenging cascade events,

the energy-dependent angular resolution of the sample is improved over the previous selection

S9
 5000
A
Atmo µ Astro ν Data
4000 Atmo ν Total ν and µ Simulation

3000
Events

2000

1000

0
1.5 B
Data/MC

1.0
0.5
−1.00 −0.75 −0.50 −0.25 0.00 0.25 0.50 0.75 1.00
sin(δ)

4000
C
Atmo µ Astro ν Data
3000 Atmo ν Total ν and µ Simulation
Events

2000

1000

0
1.5 D
Data/MC

1.0
0.5
0 5 10 15 20 25 30 35 40
Angular Uncertainty [◦ ]

Figure S2: Comparison between the data and Monte Carlo simulations. Distributions of
IceCube data and Monte Carlo simulation (MC) (42) are compared for the reconstructed dec-
lination (δ) (A) and the estimated, per-event angular uncertainty (C). Individual contributions
to the simulation are illustrated as a stacked histogram with the atmospheric muon (Atmo µ)
atmospheric neutrino (Atmo ν) and astrophysical neutrino (Astro ν) components being added
to reach the total of simulated signal and background events (red). The astrophysical flux (As-
tro ν) is an isotropic power-law model fitted to IceCube data (22), with no additional Galactic
component included. Panels B and D show the ratio of experimental data and simulation, in-
cluding statistical 1σ uncertainties from experimental data and limited Monte Carlo statistics.
Systematic uncertainties are not included.

S10
 107 103
A This Work B Previous Cascade Analysis (12)
Reconstructed Energy Ereco / GeV

106
102

Number of Events
105

104 101

103
100
500 400 300 200 100 0 100 200 500 400 300 200 100 0 100 200
Distance to Detector Boundary / m Distance to Detector Boundary / m

Figure S3: Distribution of event vertex. The distribution of experimental data events in recon-
structed energy (Ereco ) and distance to the detector boundary are compared for this work (A)
and the previous cascade selection (12) (B). Negative distance values indicate that the interac-
tion vertex is reconstructed inside the instrumented volume and thus these events are referred to
as contained events.

Figure S4: Typical cascade event views. Example event views of a “contained” (A) and an
“uncontained” (B) cascade event. The colored blobs indicate DOMs that registered light, where
the size of these blobs corresponds to the amount of observed light and the color indicates the
time of the first registered light from early (white) to late (dark blue) times. Events such as there
have time scales of 100s of ns to a couple of µs. The magenta sphere and grey arrow show the
reconstructed vertex and direction of the neutrino interaction.

S11
 40°
80% This work 50% This work (all events)
35° 50% This work 50% This work (contained events)
30° 20% This work 50% Previous Cascade Analysis (12)
25°
Opening Angle

20°
15°
10°
5°
0°
103 104 105 106
Neutrino Energy E / GeV

Figure S5: Cascade event angular resolution. The angular resolution, defined as quantiles of
the distribution of opening angles (∆Ψ) between true and reconstructed directions, as a function
of neutrino energy (Eν ) is shown for simulated events in this work (solid, black line and shaded
regions) and the previous cascade selection (12) (dashed-dotted). The dashed, orange curve
shows the angular resolution of contained events. Systematic uncertainties are not included.

as demonstrated in Figure S5. This is accomplished by the hybrid reconstruction method (16),

which exploits more information than the CNN-based method (15, 48) used in the previous
cascade selection. The energy resolution of this sample is illustrated in in Figure S6.

Combining maximum-likelihood with deep learning

The hybrid reconstruction method is a likelihood-based reconstruction algorithm that utilizes

deep learning to approximate the underlying probability density function (PDF), i.e. the pulse
arrival time distribution at each of the 5160 DOMs for any given light emitter-receiver con-
figuration. In previous reconstruction methods (19, 29), this PDF was incorporated by di-

mensionality reductions and other approximations. Our hybrid method uses neural networks
to model these high-dimensional and complex dependencies. It is constructed to exploit the
available physical symmetries and domain knowledge. Details on how the neural network ar-

S12
 A C Charged-Current

Reconstructed Energy Ereco / GeV

Reconstructed Energy Ereco / GeV

Density (column-normalized)
Density (column-normalized)
106 106
10 1 10 1

105 105
10 2 10 2

104 104

103 10 3
103 10 3

20%
100%
B D

/Edep
50% 10%
/E

0% 0%
103 104 105 106 103 104 105 106
Neutrino Energy E / GeV Deposited Energy Edep / GeV

Figure S6: Cascade event energy resolution. The energy resolution of simulated events is
shown for the neutrino energy (A-B) and for the deposited energy of charged-current electron
neutrinos (C-D). The reconstruction method aims to infer the true deposited energy (Edep ) as
shown in Panels C and D, which is a lower bound to the true neutrino energy (Eν ) (19). Panels
A and C illustrate the correlation between reconstructed (Ereco ) and true quantities. Panels B
and D show the relative resolution, defined as the 68th percentile of the absolute values of the
relative residuals |(Etrue − Ereco )/Etrue |, where Etrue is either Eν (B) or Edep (D).

chitecture was constructed, the training procedure, and the model’s performance are provided
elsewhere (16).

Due to the DOM’s largely linear response to light intensity, any event hypothesis in Ice-
Cube can be deconstructed as a superposition of individual energy losses. The hybrid method
makes use of this fact: only a single neural network (NN) needs to be trained to perform this

elementary mapping. Complex event hypotheses can then be constructed from the superposi-
tion of multiple inference steps of this elementary NN. Since this analysis is focused on cascade
events, most events are well modeled by a single energy deposition. However, fluctuations in

energy depositions along the particle shower as well as incoming/outgoing particles and coin-
cident events require a more complex event hypothesis. To this purpose, three event hypotheses
are defined and reconstructed: a single energy deposition, two causally connected energy depo-

sitions, and two independent energy depositions. Of these three reconstructed hypotheses, the

S13
reconstruction result from the model hypothesis with the lowest estimated angular uncertainty
is chosen for each event.

The estimated per-event angular uncertainty for each of these reconstructed event hypothe-
ses is obtained by three independent, fully-connected NNs. These NNs are trained via a von
Mises-Fisher likelihood (49) to estimate the circularized angular uncertainty on each of these

three reconstructions using high-level features such as the reconstructed event properties, the
difference in log-likelihood values between the hypotheses, and the second derivatives of the
likelihood. This method provides better uncertainty estimation than by evaluating the Hessian

of the likelihood at the best-fitting position. More sophisticated scans of the high-dimensional
likelihood landscape were not performed due to computational constraints. This analysis uti-
lizes circularized angular uncertainties, symmetrical in right ascension and declination. This

assumption simplifies the analysis tools used for source searches, but it is not fully valid be-
cause cascades, in particular, can have asymmetric uncertainty contours, often elongated in
right ascension due to the detector layout.

Systematic uncertainties in the detector properties have an energy-dependent impact on the

angular resolution, further degrading the resolution shown in Figure S5 by about 5% at 1 TeV
and about 30% at 1 PeV. These systematic uncertainties encompass properties of the detec-

tor medium including absorption and scattering coefficients, ice anisotropy, and the acceptance
parameterization of the refrozen ice column surrounding the DOMs, which has increased scat-
tering and absorption properties due to enclosed air bubbles and dust impurities, as well as the

DOM quantum efficiency. To account for these systematic effects, the estimated angular un-
certainty is further corrected by an energy-dependent scaling factor that is determined from a
systematic dataset that continuously samples from the estimated systematic uncertainties (50).

The coverage of the resulting uncertainty estimate is compared in Figure S7 for the baseline
MC simulation and the simulation with varied systematic parameters. In the baseline simula-

S14
 100 % 100 %
80 %
A Baseline MC 80 %
B Systematic MC
True quantile

True quantile
60 % 60 %
40 % Perfect Coverage 40 % Perfect Coverage
E = 1 TeV E = 1 TeV
20 % E = 10 TeV 20 % E = 10 TeV
E = 500 TeV E = 500 TeV
0% 0%
0% 20 % 40 % 60 % 80 % 100 % 0% 20 % 40 % 60 % 80 % 100 %
Estimated quantile Estimated quantile

Figure S7: Angular uncertainty estimator coverage. The coverage of the angular uncertainty
estimator is evaluated for different neutrino energies (colored lines) in the baseline MC simula-
tion (A) and in a simulation with varied systematic parameters (50) (B). The angular uncertainty
estimator provides an estimate for the fraction of events for which the true direction lies within
a certain quantile (x-axis). This is compared to the true frequency at which the true direction
falls into that quantile (y-axis). Perfect coverage would fall along the diagonal indicated by the
gray, dashed line. Curves that lie above this diagonal over-cover the true uncertainty, i.e. the
estimated uncertainty is larger than the true uncertainty of the reconstruction method.

tion, the uncertainty estimate is over-covering by about 15% due to the applied scaling factor.
This over-coverage is reduced when introducing systematic uncertainties in the detector proper-
ties (Figure S7B). In both cases, the coverage does not strongly depend on the neutrino energy.

While an accurate uncertainty estimate is important, the computed p-values in this analysis are
insensitive to this accuracy, and independent of any mis-modeling due to the employed data-
driven search method (see below).

Spatial distribution of Galactic Models

We use three models as templates for our diffuse neutrino search of the Galactic plane. The

Galactic plane models are based on spatial and spectral information derived from previous anal-
yses of gamma-ray emission.

The π 0 diffuse neutrino model is based on the π 0 gamma-ray component of the GAL-
PROP (51) output files available from the supplemental material for the S SZ 4R 20T 150C 5

S15
model (1). The π 0 gamma-ray component is then normalized to serve as a spatial probability
template only, as in previous IceCube analyses (11, 12).

The KRAγ models (33) are based on the same underlying gamma-ray observations as the
π 0 model, but include a radially dependent diffusion coefficient for the cosmic-ray propagation,
using the cosmic-ray transport software package DRAGON (52) and a diffuse gamma-ray sim-

ulation code G AMMA -S KY (53). The files contain all-flavor neutrino flux simulations with a
cosmic-ray cutoff of 5 PeV and 50 PeV (33, 54). They were binned over the entire sky and en-
ergy range of the analysis and used as spatial and spectral templates. These files are identical to

those used in previous IceCube analyses (11,12). Here, the model flux refers to the per-neutrino
flavor, neutrino and anti-neutrino combined flux for each model.
Figure S9 illustrates the Galactic template models after convolving with detector accep-

tance (Figure S9A and Figure S9D) and additional Gaussian smearing of 7◦ (Figure S9B and
Figure S9E) and 15◦ (Panels Figure S9C and Figure S9F). At these resolutions, small scale
structures visible in the spatial emission templates are washed out by the cascade directional

resolution. This analysis is therefore robust against mis-modeling of the spatial signal compo-
nent of the Galactic emission models, by amounts up to a few degrees. While this smearing is
beneficial for the detection of unknown or poorly characterized extended sources, such as dif-

fuse emission from the Galactic plane, it complicates further dissection of the signal observed
in the emission region.
The Galactic template searches fit each model to the integrated neutrino flux, taking the

entire sky into consideration. The flux corresponding to a certain region of the sky is obtained
by scaling the sky-integrated, best-fitting flux, illustrated in Figure 5, according to the relative
contribution of that region based on the model template. Figure S8 shows such a comparison for

three different Galactic plane regions together with gamma-ray measurements from the Tibet
Air Shower Array (37) and neutrino predictions derived from gamma-ray measurements (55).

S16
The best-fitting neutrino fluxes do not constitute independent measurements for the specified
sky regions, but rather a different presentation of the sky-integrated results. This comparison

indicates that the observed neutrino flux, when interpreted as a Galactic diffuse flux, is consis-
tent with the flux level inferred by gamma-ray observatories in the TeV-PeV range. However,
contributions to the observed neutrino flux by unresolved sources cannot be constrained by our

analysis.
Due to spatial overlap of all tested neutrino emission models, combined with the large angu-
lar uncertainty of the cascade events, correlations between the results of these model searches

are expected.
Figure S10 shows the pre-trial significance map of the all-sky search overlaid with the
Galactic stacking catalog sources locations and Fermi Bubbles templates used in those source

searches. Individual event excesses are not statistically significant in excess of the background
fluctuations as shown in Table 1. However, there is an accumulation of warm spots along the
Galactic plane and near the Galactic Center region, which is also densely populated with nu-

merous sources in the catalog stacking searches. The Galactic Center is also where the bulk of
the Galactic diffuse emission is predicted by the π 0 and KRAγ templates (Figure S9).

S17
 10 6
A
E2 dEdN [GeV s 1 cm 2 sr 1] E2 dEdN [GeV s 1 cm 2 sr 1] E2 dEdN [GeV s 1 cm 2 sr 1]

KRA5 Best-Fit Flux IceCube all-sky flux (22)

KRA50 Best-Fit Flux Tibet AS+MD [E equivalent] (37)
0 Best-Fit Flux
10 7

10 8

25 < l < 100 , |b| < 5

10 9

10 6
B KRA5 Best-Fit Flux IceCube all-sky flux (22)
KRA50 Best-Fit Flux Tibet AS+MD [E equivalent] (37)
0 Best-Fit Flux
10 7

10 8

50 < l < 200 , |b| < 5

10 9

10 6
C KRA5 Best-Fit Flux Prediction of diffuse Galactic flux (55)
KRA50 Best-Fit Flux IceCube all-sky flux (22)
0 Best-Fit Flux
10 7

10 8

0 < l < 360 , |b| < 5

10 9
103 104 105 106
E [GeV]

Figure S8: Comparison between the best-fitting flux normalizations of the Galactic plane
models. Same as Figure 5, but for flux averaged over three different regions of the sky. The
average flux values are obtained by multiplying the total, sky-integrated neutrino flux from
Table 1 and Figure 5 with the relative template contribution from each region, as indicated
in the lower left of each panel. These fluxes are therefore not independent measurements in
these parts of the sky, but an alternative presentation of the sky-integrated values. Panels A-
B include gamma-ray measurements from the Tibet Air Shower Array (37) (black asterisks),
converted to a neutrino flux assuming a hadronuclear (pp) scenario (56–58) neglecting gamma-
ray attenuation. Panel C also shows a prediction for the diffuse Galactic neutrino flux (55)
(checkered area), derived from gamma-ray measurements.

S18
 0 Template
30°
15°
b = 0°
-15°
-30°
30°
15°
b = 0°
-15°
-30°
30°
15°
b = 0°
-15°
-30°
180° 120° 60° l = 0° -60° -120° -180°
KRA5 Template
30°
15°
b = 0°
-15°
-30°
30°
15°
b = 0°
-15°
-30°
30°
15°
b = 0°
-15°
-30°
180° 120° 60° l = 0° -60° -120° -180°

0.0 0.3 0.6 0.9 1.2

Density / sr 1

Figure S9: Neutrino emission models used as templates in the Galactic plane search. The
spatial templates for the π 0 (A-C) and KRA5γ (D-F) models of diffuse Galactic neutrino emis-
sion are shown. Each panel shows the Galactic plane in a band of ±30◦ in latitude (b) and
±180◦ longitude (l) in Galactic coordinates. The models are first convolved with the IceCube
detector acceptance (A, D) and then smeared with a Gaussian corresponding to the event uncer-
tainty. Two example analysis templates are shown for a smearing of 7◦ (B, E) and 15◦ (C, F).
The spatial distribution of the KRA50 5
γ model is similar to the KRAγ one shown here and it is
available in the IceCube data archive.

S19
Searches for Neutrino Emission
Maximum Likelihood-Based Searches

The search for neutrino sources, both point sources and extended, is based on the maximum-

likelihood technique used in previous analyses of IceCube data (31). The signal hypothesis
consists of neutrino emission from a particular source, following a power-law spectrum E −γ

with spectral index γ. A likelihood function L is defined to represent the different hypotheses
(31),
N
Y ns ns
L(ns , γ) = Si (γ, δi , αi , σi , Ei ) + (1 − )Bi (δi , Ei ), (1)
i=1
N N
where ns is the number of signal events, N is the total number of events, Si is the signal PDF of
the ith event, which depends on declination (δi ), right ascension (αi ), event angular uncertainty

(σi ), reconstructed energy (Ei ), and spectral index (γ); Bi is the per event background term
which depends only on the declination and energy of each event.
Previous IceCube analyses, such as (20), assumed that the data in each declination/zenith

range are background-dominated and the total event rate is a good approximation to the back-
ground. However, this assumption is not necessarily valid for our dataset. We therefore used

an alternative method, described previously (11). This introduces a data-driven PDF, D̃i , which
does not correspond to pure background Bi , but includes the expected contribution from the
signal PDF averaged over right ascension, S̃i , in proportion to the strength of the signal that is

being fitted:
ns ns
D̃i = (1 − )Bi + S̃i . (2)
N N
The likelihood equation is then rearranged to arrive at the “signal-subtracted" likelihood
function:
N
Y ns ns
L(ns , γ) = Si (γ, δi , αi , σi , Ei ) + D̃i (δi , Ei ) − S̃i (δi , Ei ) (3)
i=1
N N

S21
The test statistic τ is defined by the ratio of the maximized likelihood and the likelihood for the
null hypothesis:
L(ns = 0)
τ = −2 ln (4)
L(n̂s , γ)
where n̂s is the best-fitting number of signal events.
This test statistic is calculated from the experimental data and converted to a p-value by
comparing the experimental result to the test statistic distribution obtained from background

mock-experiments generated from scrambled experimental data. To prevent bias, the recon-
structed direction and energy of all events was kept blind and not considered during the de-
velopment and characterization of these or subsequent analyses. Due to IceCube’s location

at the South Pole and the rotation of Earth, backgrounds in this analysis (atmospheric events
and an assumed-isotropic astrophysical neutrino background) are uniformly distributed in right
ascension. Similarly, the impact of any terrestrial sources of systematic uncertainty is also dis-

tributed evenly in right ascension. The data-driven search method makes use of this, generating
background mock-experiments by randomizing the right ascension values of the observed ex-
perimental data. This randomization removes signal from sources by evening out anisotropic

structure on the sky (in given declination bands), while maintaining the influence of system-
atic uncertainties and the statistical properties of the background events in the experimental
data. This data-driven procedure was chosen to reduce the risk of false discovery of anisotropic

structures on the sky.

To account for multiple hypotheses tested in one analysis, such as the three similar Galactic
plane emission templates, we convert the most significant of the p-values into a post-trial p-

value by using a trial-factor. This trial factor is equal to the number of tests for uncorrelated
hypotheses, but can be lessened by taking correlations into account.

S22
Signal Hypotheses

We separate signal hypotheses into two components, SSpace and SEnergy . These are defined per
event and rely on the observables of reconstructed direction, energy, and angular uncertainty.

SEnergy exploits the different energy spectrum for astrophysical neutrinos relative to that of the
background of atmospheric neutrinos and muons. In general, higher energy events provide
more support for the signal hypothesis, because the expected Galactic-plane neutrino flux has a

spectrum that falls less steeply with energy than that of the atmospheric backgrounds.
SSpace is defined based on the expected signal hypothesis. For a point source, SSpace is given
for each event by a von Mises-Fisher distribution based on the event’s reconstructed direction

(xi ), its estimated angular uncertainty (σi ) and the source location (xs ).
For template searches, full sky templates are binned in equal-solid-angle bins, convolved
with the detector acceptance, and smeared with a Gaussian with width equal to each event’s

angular uncertainty. Each event therefore has a particular spatial weight based on its position,
energy, direction, estimated angular uncertainty, and the signal hypothesis under investigation.
These combined spatial and energy weights are multiplied and the likelihood is maximized for

the analysis parameters described above.

Effective Area

The effective area is defined as the declination and energy dependent area of a 100% efficient
detector and is calculated using simulations. We calculate the effective areas as the sum of

all neutrino flavors and types assuming equal flavor ratios. Effective areas in Figure 2 are
presented as integrated over declination ranges and then divided over the solid angle to produce
an averaged effective area for that region. For a given flux Φ, which is equal to both the neutrino
dN
and anti-neutrino fluxes, Φ = Φν = Φν̄ , the rate of observed neutrinos in time, dt
, is given as:

S23
 ∞
dN
Z Z
= dΩ Aν+ν̄
eff (Eν , δ) × Φ(Eν ) dEν , (5)
dt 0

where Aeff is the energy and declination dependent effective area and Ω denotes the solid angle.

Systematic Uncertainties and their Impact

Here we consider the impact of systematic errors on our results. The data-driven search method
(see above) returns p-values that are derived from scrambled experimental data, thereby reduc-
ing the impact of un-modelled systematic uncertainties. The performance of the event selection

and reconstruction methods has been validated using simulation samples (42). The projected
analysis sensitivities and effective areas, used to convert the fitted number of events ns to a
flux normalization, are based on simulations of the detector (42). Therefore the estimated flux

normalization is affected by systematic uncertainties, while the p-value remains robust.

Systematic uncertainties pertaining to ice-modeling (scattering, absorption, anisotropy, and
properties of refrozen hole-ice column) as well as detector parameters such as the DOM quan-

tum efficiency are accounted for in multiple places. The CNNs in the event selection pipeline
are trained on a variety of different simulation datasets with different systematic properties (50).
The impact of these systematics on the CNNs has been quantified by previous work (15). The

neural network model utilized in the hybrid method is also trained on simulations that sample
different sets of systematic parameters for each subset of events (50). For the loss function,
used to evaluate prediction errors during training of the NN, the model utilizes a likelihood that

incorporates the over-dispersion resulting from marginalization of systematic parameters (16).

The event reconstruction (see above) increases the per-event angular uncertainty estimates to ac-
count for these sources of systematic uncertainties. The confidence interval construction of the

best-fitting flux normalization of the Galactic plane models, reported in Table 1 and Figure 5,
also account for these systematic uncertainties. The confidence intervals are constructed by

S24
inversion of the likelihood ratio test (59) while also sampling a realisation of systematic param-
eters for every trial. As such, the confidence intervals not only include statistical uncertainties,

but also known sources of systematic uncertainties.

We find this set of systematic uncertainties impact the sensitivity by up to 20% and the effec-
tive area by about 10%. The largest impact is on the angular resolution of high-energy events,

which are most affected by systematic uncertainties in the modeling of the glacial ice. Each of
these effects are – by construction – accounted for in the p-value calculation. Estimates of the
best-fitting flux normalizations are susceptible to systematic uncertainties, but the identification

of neutrino emission from the Galactic plane is robust due to the data-driven search method.

S25
Table S1: All-sky search most significant locations. Results of the search for the most signif-
icant point in the Northern and Southern hemisphere. Best-fitting number of signal events (ns ),
spectrum (γ) pre-trial p-value (p-valuepre ), as well as the trial corrected p-value (p-valuepost )
are presented. Other points in the sky are almost as significant.
Analysis α[◦ ] δ[◦ ] ns γ p-valuepre p-valuepost
Hotspot (North) 337.9 17.6 213.7 3.6 3.92×10−4 0.28
Hotspot (South) 248.1 -50.9 90.2 2.9 1.31×10−3 0.46

Supplemental Text
Additional Neutrino Source Searches
All-Sky Scan

A test for point source neutrino emission is performed over the full sky, searching for an excess
of neutrino events over the background. The method matches previous searches (12, 20) where

all directions in the sky are evaluated as a potential point source, by fitting the number of signal
events ns and the power-law spectral index γ. This test is evaluated on a grid of points of equal
solid angle bins spaced 0.45◦ apart, looking for the brightest spot in each hemisphere. Although

individual points are highly correlated with their neighbors, this test still entails a large (∼500)
trial factor due to searching the whole sky. The final results are reported in Table S1 as the trial-
corrected p-value (p-valuepost ) for the hottest point in the Northern and Southern Hemisphere.

Both results are consistent with the background-only hypothesis.

A skymap illustrating the best-fitting spectral index and significance at each location in the
sky is provided in Figure S11. The all-sky map shows some warm spots in coincidence with

nearby known gamma-ray sources, such as the Crab Nebula. However, for the point-source tests
that were defined a priori (all-sky scan and source list search), no individual point-like source is
statistically significant after accounting for the trial factor for the corresponding analysis. The

ability to spatially resolve a neutrino source in the all-sky scan depends on the expected number
of signal events and their energy distribution. Due to the typical 5◦ to 20◦ angular uncertainty

S26
Table S2: Fermi Bubble neutrino searches. Pre-trial significance, sensitivity, best-fitting num-
ber of signal events (ns ), and 90% Upper Limits (UL) for Fermi Bubbles model with various
exponential cutoffs. The per-flavor neutrino flux sensitivity and upper limits are given as E2 dN
dE
at 1 TeV in units of 10−12 TeV cm−2 s−1 .
Cutoff Sensitivity Φ ns p-value Significance (σ) UL Φ
50 TeV 20.87 96.0 0.03 1.88 <52.51
100 TeV 14.68 70.9 0.05 1.65 <34.06
500 TeV 7.74 34.4 0.12 1.17 <14.99
No Cutoff 3.15 23.7 0.14 1.06 <5.91

of individual cascade events, the combination of many such events is required to improve the

ability to detect a source in the sky. Assuming the source is a point source with parameter values
as measured, we expect to resolve the source to a few degrees.

Fermi Bubbles and Source Stacking

Neutrino emission from the Fermi Bubbles region around the Galactic Center was searched

for. The basic template used matches that in previous searches of IceCube data (12), and is
constructed by creating two circular lobes of radius 25◦ tangent to each other at the Galactic
center. Each point in those lobes is assumed equally likely to emit neutrinos following an E −2

spectrum, with different spectral cutoffs being tested. The full results are shown in Table S2,
and are all consistent with background. Multiple tests were performed and a trial corrected p-
value of 6.48 × 10−2 (1.52σ) is obtained that accounts for the correlations between the different

cutoffs, and this value is reported in Table 1.

The full results for stacking catalogs are shown in Table S3. As described in the main text,
the analysis searches for total emission from a catalog of point-like sources. Each source in a

particular catalog is weighted to equally contribute to the flux before any detector effects are
considered. The analysis determines the best-fitting total number of signal events and therefore
total flux and a single catalog power law spectral index (γ).

S27
Table S3: Source catalog search summary results. Pre-trial significance, sensitivity, best-
fitting spectrum (γ), total number of signal events (ns ), flux, and 90% Upper Limits (UL) for the
Galactic stacking catalog analyses. Upper limits are with respect to a source emitting following
an E−2 spectrum. The per-flavor neutrino flux sensitivity, best-fit, and upper limits are given as
E2 dNdE
at 100 TeV in units of 10−12 TeV cm−2 s−1 for the entire catalog of sources.
Catalog Sensitivity Φ ns γ p-value Signficance (σ) Flux Φ UL Φ
SNR 2.24 218.6 2.75 5.90×10−4 3.24 6.22 <9.01
PWN 2.25 279.6 3.00 5.93×10−4 3.24 3.80 <9.50
UNID 1.89 238.4 2.85 3.39×10−4 3.40 5.03 <7.76

Source List

A search was performed using a list of 109 a priori selected positions in the sky. The source
list was constructed based on the GeV gamma-ray flux (60) catalog with previously described
methods (20) but optimized for the declination dependent sensitivity of this event selection. The

list consists mainly of extragalactic objects. A point source likelihood test is performed at each
position, and the results are shown in Table S5. All results are consistent with background, and
90% confidence flux upper limits are placed on the source flux Φ at Eν = 100 TeV assuming

an E −2 (Φ2 ) or E −3 (Φ3 ) spectrum. Upper limits for under-fluctuations are shown at the point
source sensitivity by convention. The declination dependent sensitivity is shown in Figure S12
along with upper limits for each spectrum. This work is an improvement to the sensitivity in

the Southern Sky, however track-based analyses are still much more sensitive in the Northern
Sky and individual point sources there are expected to emerge in track analyses before cascade
analyses.

S29
Table S4: Galactic source catalog. The sources and locations (34, 35) of galactic sources used
in stacking catalogs.
Catalog Source α [◦ ] δ [◦ ]
SNR Vela Junior 133.0 -46.33
SNR RX J1713.7-3946 258.36 -39.77
SNR HESS J1614-518 243.56 -51.82
SNR HESS J1457-593 223.7 -59.07
SNR SNR G323.7-01.0 233.63 -57.2
SNR HESS J1731-347 262.98 -34.71
SNR Gamma Cygni 305.27 40.52
SNR RCW 86 220.12 -62.65
SNR HESS J1912+101 288.33 10.19
SNR HESS J1745-303 266.3 -30.2
SNR Cassiopeia A 350.85 58.81
SNR CTB 37A 258.64 -38.54
PWN Vela X 128.29 -45.19
PWN Crab nebula 83.63 22.01
PWN HESS J1708-443 257.0 -44.3
PWN HESS J1825-137 276.55 -13.58
PWN HESS J1632-478 248.01 -47.87
PWN MSH 15-52 228.53 -59.16
PWN HESS J1813-178 273.36 -17.86
PWN HESS J1303-631 195.75 -63.2
PWN HESS J1616-508 244.06 -50.91
PWN HESS J1418-609 214.69 -60.98
PWN HESS J1837-069 279.43 -6.93
PWN HESS J1026-582 157.17 -58.29
UNID MGRO J1908+06 286.91 6.32
UNID Westerlund 1 251.5 -45.8
UNID HESS J1702-420 255.68 -42.02
UNID 2HWC J1814-173 273.52 -17.31
UNID HESS J1841-055 280.23 -5.55
UNID 2HWC J1819-150 274.83 -15.06
UNID HESS J1804-216 271.12 -21.73
UNID HESS J1809-193 272.63 -19.3
UNID HESS J1843-033 280.75 -3.3
UNID TeV J2032+4130 307.93 41.51
UNID HESS J1708-410 257.10 -41.09
UNID HESS J1857+026 284.30 2.67

S30
 dN/dE ∝ E−2 dN/dE ∝ E−3
A B
This Work Sensitivity
E 2 · d N /d E at 100 TeV [TeV s−1cm−2]

This Work 4σ DP
10−10 Cascades (12) 10−10
Tracks (20)
ANTARES
Upper Limit 90% C.L.
10−11 10−11

10−12 10−12

10−13 10−13

−1.0 −0.5 0.0 0.5 1.0 −1.0 −0.5 0.0 0.5 1.0
sin(δ) sin(δ)

Figure S12: Source list sensitivity and upper limits Sensitivity to sources emitting an E−2
spectrum (A) and E−3 spectrum (B) for each data set. Individual sources in the source catalog
are shown with their 90% confidence level (CL) upper limits assuming an E−2 (A) and E−3 (B)
emission spectra. ANTARES results are for E−2 (61) and E−3 (62) sensitivities. We also show
previous results from IceCube tracks (20) and cascades (12). Also shown in the 4σ discovery
potential (DP) for this work. All results are consistent with background.

S31
Table S5: Source list search results. Sources in the source-list analysis, location (60), and
pre-trial p-values. Per-flavor 90% flux upper limits are shown as E2 dN
dE
at 100 TeV in units of
10−12 TeV cm−2 s−1 for sources emitting following an E−2 (UL Φ2 ) and E−3 (UL Φ3 ) spectrum.

Source Name α [◦ ] δ [◦ ] ns γ p-value UL Φ2 UL Φ3

1 G343.1-2.3 257.0 -44.3 77.6 2.9 0.011 <3.34 <1.58
2 HESS J0835-455 128.3 -45.2 0.2 2.0 0.719 <0.59 <0.28
3 PKS 0426-380 67.2 -37.9 3.0 2.3 0.640 <0.81 <0.42
4 PKS 2155-304 329.7 -30.2 61.5 4.0 0.249 <1.52 <1.02
5 Mkn 421 166.1 38.2 0.0 – – <1.13 <0.32
6 PKS 0537-441 84.7 -44.1 1.1 1.8 0.576 <0.89 <0.45
7 PKS 0447-439 72.4 -43.8 0.7 2.1 0.698 <0.62 <0.31
8 BL Lac 330.7 42.3 7.6 1.6 0.340 <2.73 <0.77
9 PG 1553+113 238.9 11.2 0.0 – – <0.68 <0.35
10 TXS 0518+211 80.4 21.2 149.3 3.4 0.035 <4.24 <1.59
11 PKS 0235+164 39.7 16.6 0.0 – – <0.67 <0.31
12 PKS 1424+240 216.8 23.8 7.7 1.3 0.095 <3.55 <1.26
13 3C 66A 35.7 43.0 85.8 3.7 0.238 <3.44 <0.89
14 TXS 0506+056 77.4 5.7 0.0 – – <0.61 <0.39
15 AP Librae 229.4 -24.4 0.3 1.0 0.725 <0.61 <0.34
16 S5 0716+71 110.5 71.3 1.9 1.0 0.532 <3.94 <0.58
17 B2 1215+30 184.5 30.1 5.2 2.2 0.573 <1.80 <0.53
18 MH 2136-428 324.9 -42.6 0.0 – – <0.53 <0.27
19 PKS 2233-148 339.1 -14.6 25.8 4.0 0.584 <0.93 <0.65
20 Mkn 501 253.5 39.8 4.1 1.0 0.514 <2.09 <0.60
21 PMN J1603-4904 241.0 -49.1 84.5 3.1 0.010 <3.49 <1.59
22 S2 0109+22 18.0 22.8 147.8 4.0 0.077 <3.75 <1.31
23 PKS 0301-243 45.9 -24.1 44.3 3.6 0.405 <1.37 <0.76
24 4C +01.28 164.6 1.6 16.1 1.2 0.025 <3.27 <2.42
25 PKS 0700-661 105.1 -66.2 28.8 2.9 0.239 <1.72 <0.82
26 TXS 0628-240 97.7 -24.1 0.9 3.0 0.746 <0.55 <0.31
27 PKS 0823-223 126.5 -22.5 2.0 2.0 0.553 <1.01 <0.58
28 PKS 0735+17 114.5 17.7 63.6 4.0 0.376 <2.16 <0.91
29 PMN J1329-5608 202.3 -56.1 40.3 4.0 0.276 <1.48 <0.77
30 PMN J0531-4827 83.0 -48.5 0.0 – – <0.55 <0.26
31 MG1 J021114+1051 32.8 10.9 28.7 4.0 0.622 <1.22 <0.66
32 PKS 1440-389 221.0 -39.1 11.0 1.0 0.175 <2.07 <1.10
33 OT 081 267.9 9.6 35.2 2.8 0.356 <1.88 <1.10
34 OJ 287 133.7 20.1 28.6 2.7 0.473 <1.84 <0.74
35 PKS 1101-536 166.0 -54.0 0.0 – – <0.46 <0.23
36 TXS 0141+268 26.2 27.1 1.4 3.2 0.706 <1.17 <0.37

S32
 Source Name α [◦ ] δ [◦ ] ns γ p-value UL Φ2 UL Φ3
37 1H 1013+498 153.8 49.4 4.2 2.6 0.676 <1.65 <0.36
38 PKS 0048-09 12.7 -9.5 115.2 3.6 0.089 <2.35 <1.68
39 PMN J1650-5044 252.6 -50.8 88.1 2.9 0.002 <4.02 <1.93
40 PKS 0118-272 20.1 -27.0 44.8 3.9 0.378 <1.38 <0.82
41 1H 1914-194 289.4 -19.4 26.9 2.5 0.152 <2.30 <1.26
42 PKS 0332-403 53.6 -40.1 9.8 2.4 0.481 <1.13 <0.59
43 OJ 014 122.9 1.8 2.6 4.0 0.771 <0.57 <0.39
44 PMN J1918-4111 289.6 -41.2 54.6 3.7 0.256 <1.72 <0.89
45 PKS 1936-623 295.3 -62.2 26.6 4.0 0.420 <1.19 <0.55
46 1H 1720+117 261.3 11.9 6.8 2.7 0.712 <0.91 <0.49
47 PMN J1610-6649 242.7 -66.8 10.1 2.9 0.588 <0.84 <0.41
48 PMN J0334-3725 53.6 -37.4 10.7 2.4 0.456 <1.25 <0.65
49 TXS 1714-336 259.4 -33.7 67.7 2.9 0.028 <2.68 <1.54
50 PKS 2005-489 302.4 -48.8 0.0 – – <0.54 <0.26
51 PKS B1056-113 164.8 -11.6 19.6 2.1 0.096 <2.27 <1.66
52 RGB J2243+203 341.0 20.4 163.1 4.0 0.022 <4.53 <1.82
53 1ES 1959+650 300.0 65.1 150.0 4.0 0.071 <7.30 <1.14
54 S4 0814+42 124.6 42.4 15.4 4.0 0.568 <1.74 <0.49
55 KUV 00311-1938 8.4 -19.4 109.8 3.6 0.063 <2.91 <1.61
56 PMN J2250-2806 342.7 -28.1 59.1 4.0 0.244 <1.62 <1.05
57 1RXS J130421.2-435308 196.1 -43.9 7.1 2.3 0.473 <1.11 <0.57
58 PMN J0810-7530 122.8 -75.5 16.5 4.0 0.598 <0.79 <0.44
59 3C 454.3 343.5 16.2 106.2 4.0 0.188 <2.67 <1.25
60 PKS 1424-41 217.0 -42.1 10.1 1.0 0.280 <1.63 <0.83
61 3C 279 194.0 -5.8 10.1 1.0 0.120 <1.90 <1.63
62 CTA 102 338.2 11.7 129.6 4.0 0.102 <3.09 <1.60
63 PKS 1510-089 228.2 -9.1 4.8 1.0 0.212 <1.79 <1.34
64 PKS 0454-234 74.3 -23.4 0.0 – – <0.55 <0.30
65 PKS 1502+106 226.1 10.5 8.7 1.0 0.422 <1.79 <0.99
66 PKS 1830-211 278.4 -21.1 70.4 2.8 0.033 <3.07 <1.68
67 PKS 2326-502 352.3 -49.9 6.6 2.1 0.265 <1.66 <0.84
68 PKS 0727-11 112.6 -11.7 8.2 2.6 0.598 <0.83 <0.65
69 4C +21.35 186.2 21.4 11.4 1.5 0.162 <3.09 <1.15
70 PMN J2345-1555 356.3 -15.9 11.9 3.3 0.646 <0.78 <0.53
71 4C +01.02 17.2 1.6 74.8 3.8 0.336 <1.66 <1.18
72 PKS 2023-07 306.4 -7.6 122.8 3.5 0.060 <2.32 <1.76
73 Ton 599 179.9 29.2 14.5 2.4 0.405 <2.49 <0.74
74 4C +38.41 248.8 38.1 3.5 1.0 0.573 <1.84 <0.54
75 PKS 1244-255 191.7 -25.8 3.9 1.0 0.128 <2.04 <1.28
76 B2 1520+31 230.5 31.7 9.3 1.0 0.013 <6.59 <1.93

S33
 Source Name α [◦ ] δ [◦ ] ns γ p-value UL Φ2 UL Φ3
77 PKS 1124-186 171.8 -19.0 0.0 – – <0.56 <0.33
78 4C +28.07 39.5 28.8 0.0 – – <1.15 <0.36
79 PKS 1730-13 263.3 -13.1 51.2 2.7 0.128 <2.27 <1.59
80 PKS 0805-07 122.1 -7.9 2.4 2.3 0.676 <0.69 <0.53
81 PKS 0208-512 32.7 -51.0 4.9 1.8 0.144 <1.97 <1.03
82 PKS 0336-01 54.9 -1.8 11.0 4.0 0.729 <0.62 <0.47
83 PKS 0502+049 76.3 5.0 0.0 – – <0.61 <0.39
84 PKS 0402-362 61.0 -36.1 2.2 2.2 0.672 <0.71 <0.38
85 PMN J1802-3940 270.7 -39.7 76.7 3.1 0.043 <2.81 <1.39
86 PKS 1622-253 246.4 -25.5 36.6 2.6 0.089 <2.26 <1.36
87 PKS 2052-47 314.1 -47.2 0.0 – – <0.55 <0.26
88 3C 273 187.3 2.0 11.3 1.0 0.227 <2.01 <1.41
89 PKS 2142-75 326.8 -75.6 2.6 1.0 0.744 <0.49 <0.26
90 MG1 J123931+0443 189.9 4.7 9.3 1.0 0.217 <2.21 <1.42
91 MG2 J201534+3710 303.9 37.2 113.6 3.1 0.039 <5.29 <1.51
92 Galactic Centre 266.4 -29.0 58.7 2.7 0.022 <2.79 <1.67
93 PKS 2247-131 342.5 -12.8 7.3 4.0 0.709 <0.62 <0.46
94 NGC 1275 50.0 41.5 29.3 3.3 0.529 <2.01 <0.57
95 PKS 0521-36 80.7 -36.5 5.7 2.0 0.237 <1.75 <1.00
96 Cen A 201.0 -43.5 9.4 2.4 0.398 <1.30 <0.66
97 LMC 80.0 -68.8 31.5 4.0 0.327 <1.38 <0.71
98 SMC 14.5 -72.8 43.1 1.0 0.174 <1.77 <0.94
99 NGC 4945 196.4 -49.5 7.3 2.4 0.425 <1.28 <0.63
100 NGC 253 11.9 -25.3 51.4 3.8 0.328 <1.43 <0.88
101 NGC 1068 40.7 -0.0 87.1 4.0 0.252 <1.88 <1.36
102 M 82 148.9 69.7 22.9 2.9 0.564 <3.44 <0.53
103 Arp 220 233.7 23.5 15.2 1.0 0.009 <5.18 <2.03
104 M 31 10.8 41.2 18.2 2.9 0.487 <2.21 <0.62
105 NGC 3424 162.9 32.9 0.0 – – <1.19 <0.34
106 IC 678 168.6 6.6 19.2 1.8 0.005 <4.55 <2.85
107 NGC 5380 209.3 37.5 10.6 1.0 0.249 <3.30 <0.93
108 Arp 299 172.1 58.5 0.0 – – <1.25 <0.23
109 NGC 2146 94.5 78.3 4.2 1.0 0.445 <4.71 <0.73

S34
References and Notes
1. M. Ackermann, M. Ajello, W. B. Atwood, L. Baldini, J. Ballet, G. Barbiellini, D. Bastieri, K.
Bechtol, R. Bellazzini, B. Berenji, R. D. Blandford, E. D. Bloom, E. Bonamente, A. W.
Borgland, T. J. Brandt, J. Bregeon, M. Brigida, P. Bruel, R. Buehler, S. Buson, G. A.
Caliandro, R. A. Cameron, P. A. Caraveo, E. Cavazzuti, C. Cecchi, E. Charles, A.
Chekhtman, J. Chiang, S. Ciprini, R. Claus, J. Cohen-Tanugi, J. Conrad, S. Cutini, A. de
Angelis, F. de Palma, C. D. Dermer, S. W. Digel, E. do Couto e Silva, P. S. Drell, A.
Drlica-Wagner, L. Falletti, C. Favuzzi, S. J. Fegan, E. C. Ferrara, W. B. Focke, P. Fortin,
Y. Fukazawa, S. Funk, P. Fusco, D. Gaggero, F. Gargano, S. Germani, N. Giglietto, F.
Giordano, M. Giroletti, T. Glanzman, G. Godfrey, J. E. Grove, S. Guiriec, M. Gustafsson,
D. Hadasch, Y. Hanabata, A. K. Harding, M. Hayashida, E. Hays, D. Horan, X. Hou, R.
E. Hughes, G. Jóhannesson, A. S. Johnson, R. P. Johnson, T. Kamae, H. Katagiri, J.
Kataoka, J. Knödlseder, M. Kuss, J. Lande, L. Latronico, S.-H. Lee, M. Lemoine-
Goumard, F. Longo, F. Loparco, B. Lott, M. N. Lovellette, P. Lubrano, M. N. Mazziotta,
J. E. McEnery, P. F. Michelson, W. Mitthumsiri, T. Mizuno, C. Monte, M. E. Monzani,
A. Morselli, I. V. Moskalenko, S. Murgia, M. Naumann-Godo, J. P. Norris, E. Nuss, T.
Ohsugi, A. Okumura, N. Omodei, E. Orlando, J. F. Ormes, D. Paneque, J. H. Panetta, D.
Parent, M. Pesce-Rollins, M. Pierbattista, F. Piron, G. Pivato, T. A. Porter, S. Rainò, R.
Rando, M. Razzano, S. Razzaque, A. Reimer, O. Reimer, H. F.-W. Sadrozinski, C. Sgrò,
E. J. Siskind, G. Spandre, P. Spinelli, A. W. Strong, D. J. Suson, H. Takahashi, T.
Tanaka, J. G. Thayer, J. B. Thayer, D. J. Thompson, L. Tibaldo, M. Tinivella, D. F.
Torres, G. Tosti, E. Troja, T. L. Usher, J. Vandenbroucke, V. Vasileiou, G. Vianello, V.
Vitale, A. P. Waite, P. Wang, B. L. Winer, K. S. Wood, M. Wood, Z. Yang, M. Ziegler,
S. Zimmer, Fermi-LAT observations of the diffuse γ-ray emission: Implications for
cosmic rays and the interstellar medium. Astrophys. J. 750, 3 (2012).
2. F. W. Stecker, Diffuse fluxes of cosmic high-energy neutrinos. Astrophys. J. 228, 919 (1979).
3. V. Berezinsky, T. Gaisser, F. Halzen, T. Stanev, Diffuse radiation from cosmic ray
interactions in the galaxy. Astropart. Phys. 1, 281–287 (1993).
4. G. Ingelman, M. Thunman, arXiv:hep-ph/9604286 (1996).
5. C. Evoli, D. Grasso, L. Maccione, Diffuse neutrino and gamma-ray emissions of the galaxy
above the TeV. J. Cosmol. Astropart. Phys. 2007, 003 (2007).
6. M. Ackermann, M. Ajello, A. Allafort, L. Baldini, J. Ballet, G. Barbiellini, M. G. Baring, D.
Bastieri, K. Bechtol, R. Bellazzini, R. D. Blandford, E. D. Bloom, E. Bonamente, A. W.
Borgland, E. Bottacini, T. J. Brandt, J. Bregeon, M. Brigida, P. Bruel, R. Buehler, G.
Busetto, S. Buson, G. A. Caliandro, R. A. Cameron, P. A. Caraveo, J. M. Casandjian, C.
Cecchi, O. Celik, E. Charles, S. Chaty, R. C. G. Chaves, A. Chekhtman, C. C. Cheung, J.
Chiang, G. Chiaro, A. N. Cillis, S. Ciprini, R. Claus, J. Cohen-Tanugi, L. R. Cominsky, J.
Conrad, S. Corbel, S. Cutini, F. D’Ammando, A. de Angelis, F. de Palma, C. D. Dermer,
E. do Couto e Silva, P. S. Drell, A. Drlica-Wagner, L. Falletti, C. Favuzzi, E. C. Ferrara,
A. Franckowiak, Y. Fukazawa, S. Funk, P. Fusco, F. Gargano, S. Germani, N. Giglietto,
P. Giommi, F. Giordano, M. Giroletti, T. Glanzman, G. Godfrey, I. A. Grenier, M.-H.
Grondin, J. E. Grove, S. Guiriec, D. Hadasch, Y. Hanabata, A. K. Harding, M.
Hayashida, K. Hayashi, E. Hays, J. W. Hewitt, A. B. Hill, R. E. Hughes, M. S. Jackson,
 T. Jogler, G. Jóhannesson, A. S. Johnson, T. Kamae, J. Kataoka, J. Katsuta, J.
Knödlseder, M. Kuss, J. Lande, S. Larsson, L. Latronico, M. Lemoine-Goumard, F.
Longo, F. Loparco, M. N. Lovellette, P. Lubrano, G. M. Madejski, F. Massaro, M.
Mayer, M. N. Mazziotta, J. E. McEnery, J. Mehault, P. F. Michelson, R. P. Mignani, W.
Mitthumsiri, T. Mizuno, A. A. Moiseev, M. E. Monzani, A. Morselli, I. V. Moskalenko,
S. Murgia, T. Nakamori, R. Nemmen, E. Nuss, M. Ohno, T. Ohsugi, N. Omodei, M.
Orienti, E. Orlando, J. F. Ormes, D. Paneque, J. S. Perkins, M. Pesce-Rollins, F. Piron, G.
Pivato, S. Rainò, R. Rando, M. Razzano, S. Razzaque, A. Reimer, O. Reimer, S. Ritz, C.
Romoli, M. Sánchez-Conde, A. Schulz, C. Sgrò, P. E. Simeon, E. J. Siskind, D. A. Smith,
G. Spandre, P. Spinelli, F. W. Stecker, A. W. Strong, D. J. Suson, H. Tajima, H.
Takahashi, T. Takahashi, T. Tanaka, J. G. Thayer, J. B. Thayer, D. J. Thompson, S. E.
Thorsett, L. Tibaldo, O. Tibolla, M. Tinivella, E. Troja, Y. Uchiyama, T. L. Usher, J.
Vandenbroucke, V. Vasileiou, G. Vianello, V. Vitale, A. P. Waite, M. Werner, B. L.
Winer, K. S. Wood, M. Wood, R. Yamazaki, Z. Yang, S. Zimmer, Detection of the
characteristic pion-decay signature in supernova remnants. Science 339, 807–811 (2013).
7. D. Guetta, E. Amato, Neutrino flux predictions for galactic plerions. Astropart. Phys. 19, 403–
407 (2003).
8. W. Bednarek, Neutrinos from the pulsar wind nebulae. Astron. Astrophys. 407, 1–6 (2003).
9. M. C. Gonzalez-Garcia, F. Halzen, S. Mohapatra, Identifying Galactic PeVatrons with
neutrinos. Astropart. Phys. 31, 437–444 (2009).
10. F. Halzen, A. Kheirandish, V. Niro, Prospects for detecting galactic sources of cosmic
neutrinos with IceCube: An update. Astropart. Phys. 86, 46–56 (2017).
11. M. G. Aartsen, M. Ackermann, J. Adams, J. A. Aguilar, M. Ahlers, M. Ahrens, I. A.
Samarai, D. Altmann, K. Andeen, T. Anderson, I. Ansseau, G. Anton, C. Argüelles, J.
Auffenberg, S. Axani, H. Bagherpour, X. Bai, J. P. Barron, S. W. Barwick, V. Baum, R.
Bay, J. J. Beatty, J. B. Tjus, K.-H. Becker, S. BenZvi, D. Berley, E. Bernardini, D. Z.
Besson, G. Binder, D. Bindig, E. Blaufuss, S. Blot, C. Bohm, M. Börner, F. Bos, D.
Bose, S. Böser, O. Botner, J. Bourbeau, F. Bradascio, J. Braun, L. Brayeur, M. Brenzke,
H.-P. Bretz, S. Bron, A. Burgman, T. Carver, J. Casey, M. Casier, E. Cheung, D. Chirkin,
A. Christov, K. Clark, L. Classen, S. Coenders, G. H. Collin, J. M. Conrad, D. F. Cowen,
R. Cross, M. Day, J. P. A. M. André, C. D. Clercq, J. J. DeLaunay, H. Dembinski, S. D.
Ridder, P. Desiati, K. D. Vries, G. Wasseige, M. With, T. DeYoung, J. C. Díaz-Vélez, V.
Lorenzo, H. Dujmovic, J. P. Dumm, M. Dunkman, B. Eberhardt, T. Ehrhardt, B.
Eichmann, P. Eller, P. A. Evenson, S. Fahey, A. R. Fazely, J. Felde, K. Filimonov, C.
Finley, S. Flis, A. Franckowiak, E. Friedman, T. Fuchs, T. K. Gaisser, J. Gallagher, L.
Gerhardt, K. Ghorbani, W. Giang, T. Glauch, T. Glüsenkamp, A. Goldschmidt, J. G.
Gonzalez, D. Grant, Z. Griffith, C. Haack, A. Hallgren, F. Halzen, K. Hanson, D.
Hebecker, D. Heereman, K. Helbing, R. Hellauer, S. Hickford, J. Hignight, G. C. Hill, K.
D. Hoffman, R. Hoffmann, B. Hokanson-Fasig, K. Hoshina, F. Huang, M. Huber, K.
Hultqvist, S. In, A. Ishihara, E. Jacobi, G. S. Japaridze, M. Jeong, K. Jero, B. J. P. Jones,
P. Kalacynski, W. Kang, A. Kappes, T. Karg, A. Karle, U. Katz, M. Kauer, A. Keivani, J.
L. Kelley, A. Kheirandish, J. Kim, M. Kim, T. Kintscher, J. Kiryluk, T. Kittler, S. R.
Klein, G. Kohnen, R. Koirala, H. Kolanoski, L. Köpke, C. Kopper, S. Kopper, J. P.
Koschinsky, D. J. Koskinen, M. Kowalski, K. Krings, M. Kroll, G. Krückl, J. Kunnen, S.
 Kunwar, N. Kurahashi, T. Kuwabara, A. Kyriacou, M. Labare, J. L. Lanfranchi, M. J.
Larson, F. Lauber, D. Lennarz, M. Lesiak-Bzdak, M. Leuermann, Q. R. Liu, L. Lu, J.
Lünemann, W. Luszczak, J. Madsen, G. Maggi, K. B. M. Mahn, S. Mancina, R.
Maruyama, K. Mase, R. Maunu, F. McNally, K. Meagher, M. Medici, M. Meier, T.
Menne, G. Merino, T. Meures, S. Miarecki, J. Micallef, G. Momenté, T. Montaruli, R.
W. Moore, M. Moulai, R. Nahnhauer, P. Nakarmi, U. Naumann, G. Neer, H.
Niederhausen, S. C. Nowicki, D. R. Nygren, A. O. Pollmann, A. Olivas, A.
O’Murchadha, T. Palczewski, H. Pandya, D. V. Pankova, P. Peiffer, J. A. Pepper, C. P.
Heros, D. Pieloth, E. Pinat, M. Plum, P. B. Price, G. T. Przybylski, C. Raab, L. Rädel, M.
Rameez, K. Rawlins, R. Reimann, B. Relethford, M. Relich, E. Resconi, W. Rhode, M.
Richman, S. Robertson, M. Rongen, C. Rott, T. Ruhe, D. Ryckbosch, D. Rysewyk, T.
Sälzer, S. E. S. Herrera, A. Sandrock, J. Sandroos, S. Sarkar, S. Sarkar, K. Satalecka, P.
Schlunder, T. Schmidt, A. Schneider, S. Schoenen, S. Schöneberg, L. Schumacher, D.
Seckel, S. Seunarine, D. Soldin, M. Song, G. M. Spiczak, C. Spiering, J. Stachurska, T.
Stanev, A. Stasik, J. Stettner, A. Steuer, T. Stezelberger, R. G. Stokstad, A. Stößl, N. L.
Strotjohann, G. W. Sullivan, M. Sutherland, I. Taboada, J. Tatar, F. Tenholt, S. Ter-
Antonyan, A. Terliuk, G. Tešić, S. Tilav, P. A. Toale, M. N. Tobin, S. Toscano, D. Tosi,
M. Tselengidou, C. F. Tung, A. Turcati, C. F. Turley, B. Ty, E. Unger, M. Usner, J.
Vandenbroucke, W. V. Driessche, N. Eijndhoven, S. Vanheule, J. Santen, M. Vehring, E.
Vogel, M. Vraeghe, C. Walck, A. Wallace, M. Wallraff, F. D. Wandler, N. Wandkowsky,
A. Waza, C. Weaver, M. J. Weiss, C. Wendt, S. Westerhoff, B. J. Whelan, S. Wickmann,
K. Wiebe, C. H. Wiebusch, L. Wille, D. R. Williams, L. Wills, M. Wolf, J. Wood, T. R.
Wood, E. Woolsey, K. Woschnagg, D. L. Xu, X. W. Xu, Y. Xu, J. P. Yanez, G. Yodh, S.
Yoshida, T. Yuan, M. Zoll, IceCube Collaboration, Constraints on Galactic neutrino
emission with seven years of IceCube data. Astrophys. J. 849, 67 (2017).
12. M. G. Aartsen, M. Ackermann, J. Adams, J. A. Aguilar, M. Ahlers, M. Ahrens, C. Alispach,
K. Andeen, T. Anderson, I. Ansseau, G. Anton, C. Argüelles, J. Auffenberg, S. Axani, P.
Backes, H. Bagherpour, X. Bai, A. B. v, A. Barbano, S. W. Barwick, B. Bastian, V.
Baum, S. Baur, R. Bay, J. J. Beatty, K.-H. Becker, J. B. Tjus, S. BenZvi, D. Berley, E.
Bernardini, D. Z. Besson, G. Binder, D. Bindig, E. Blaufuss, S. Blot, C. Bohm, M.
Börner, S. Böser, O. Botner, J. Böttcher, E. Bourbeau, J. Bourbeau, F. Bradascio, J.
Braun, S. Bron, J. Brostean-Kaiser, A. Burgman, J. Buscher, R. S. Busse, T. Carver, C.
Chen, E. Cheung, D. Chirkin, K. Clark, L. Classen, A. Coleman, G. H. Collin, J. M.
Conrad, P. Coppin, P. Correa, D. F. Cowen, R. Cross, P. Dave, J. P. A. M. de André, C.
D. Clercq, J. J. DeLaunay, H. Dembinski, K. Deoskar, S. D. Ridder, P. Desiati, K. D.
Vries, G. Wasseige, M. With, T. DeYoung, A. Diaz, J. C. Díaz-Vélez, H. Dujmovic, M.
Dunkman, E. Dvorak, B. Eberhardt, T. Ehrhardt, P. Eller, R. Engel, P. A. Evenson, S.
Fahey, A. R. Fazely, J. Felde, K. Filimonov, C. Finley, A. Franckowiak, E. Friedman, A.
Fritz, T. K. Gaisser, J. Gallagher, E. Ganster, S. Garrappa, L. Gerhardt, K. Ghorbani, T.
Glauch, T. Glüsenkamp, A. Goldschmidt, J. G. Gonzalez, D. Grant, Z. Griffith, M.
Günder, M. Gündüz, C. Haack, A. Hallgren, L. Halve, F. Halzen, K. Hanson, A. Haungs,
D. Hebecker, D. Heereman, P. Heix, K. Helbing, R. Hellauer, F. Henningsen, S.
Hickford, J. Hignight, G. C. Hill, K. D. Hoffman, R. Hoffmann, T. Hoinka, B. Hokanson-
Fasig, K. Hoshina, F. Huang, M. Huber, T. Huber, K. Hultqvist, M. Hünnefeld, R.
Hussain, S. In, N. Iovine, A. Ishihara, G. S. Japaridze, M. Jeong, K. Jero, B. J. P. Jones,
F. Jonske, R. Joppe, D. Kang, W. Kang, A. Kappes, D. Kappesser, T. Karg, M. Karl, A.
 Karle, U. Katz, M. Kauer, J. L. Kelley, A. Kheirandish, J. Kim, T. Kintscher, J. Kiryluk,
T. Kittler, S. R. Klein, R. Koirala, H. Kolanoski, L. Köpke, C. Kopper, S. Kopper, D. J.
Koskinen, M. Kowalski, K. Krings, G. Krückl, N. Kulacz, N. Kurahashi, A. Kyriacou, M.
Labare, J. L. Lanfranchi, M. J. Larson, F. Lauber, J. P. Lazar, K. Leonard, A.
Leszczyńska, M. Leuermann, Q. R. Liu, E. Lohfink, C. J. L. Mariscal, L. Lu, F. Lucarelli,
J. Lünemann, W. Luszczak, Y. Lyu, W. Y. Ma, J. Madsen, G. Maggi, K. B. M. Mahn, Y.
Makino, P. Mallik, K. Mallot, S. Mancina, I. C. Mariş, R. Maruyama, K. Mase, R.
Maunu, F. McNally, K. Meagher, M. Medici, A. Medina, M. Meier, S. Meighen-Berger,
T. Menne, G. Merino, T. Meures, J. Micallef, G. Momenté, T. Montaruli, R. W. Moore,
R. Morse, M. Moulai, P. Muth, R. Nagai, U. Naumann, G. Neer, H. Niederhausen, S. C.
Nowicki, D. R. Nygren, A. O. Pollmann, M. Oehler, A. Olivas, A. O’Murchadha, E.
O’Sullivan, T. Palczewski, H. Pandya, D. V. Pankova, N. Park, P. Peiffer, C. P. de los
Heros, S. Philippen, D. Pieloth, E. Pinat, A. Pizzuto, M. Plum, A. Porcelli, P. B. Price, G.
T. Przybylski, C. Raab, A. Raissi, M. Rameez, L. Rauch, K. Rawlins, I. C. Rea, R.
Reimann, B. Relethford, M. Renschler, G. Renzi, E. Resconi, W. Rhode, M. Richman, S.
Robertson, M. Rongen, C. Rott, T. Ruhe, D. Ryckbosch, D. Rysewyk, I. Safa, S. E. S.
Herrera, A. Sandrock, J. Sandroos, M. Santander, S. Sarkar, S. Sarkar, K. Satalecka, M.
Schaufel, H. Schieler, P. Schlunder, T. Schmidt, A. Schneider, J. Schneider, F. G.
Schröder, L. Schumacher, S. Sclafani, D. Seckel, S. Seunarine, S. Shefali, M. Silva, R.
Snihur, J. Soedingrekso, D. Soldin, M. Song, G. M. Spiczak, C. Spiering, J. Stachurska,
M. Stamatikos, T. Stanev, R. Stein, P. Steinmüller, J. Stettner, A. Steuer, T. Stezelberger,
R. G. Stokstad, A. Stößl, N. L. Strotjohann, T. Stürwald, T. Stuttard, G. W. Sullivan, I.
Taboada, F. Tenholt, S. Ter-Antonyan, A. Terliuk, S. Tilav, L. Tomankova, C. Tönnis, S.
Toscano, D. Tosi, A. Trettin, M. Tselengidou, C. F. Tung, A. Turcati, R. Turcotte, C. F.
Turley, B. Ty, E. Unger, M. A. U. Elorrieta, M. Usner, J. Vandenbroucke, W. V.
Driessche, D. Eijk, N. Eijndhoven, S. Vanheule, J. Santen, M. Vraeghe, C. Walck, A.
Wallace, M. Wallraff, N. Wandkowsky, T. B. Watson, C. Weaver, A. Weindl, M. J.
Weiss, J. Weldert, C. Wendt, J. Werthebach, B. J. Whelan, N. Whitehorn, K. Wiebe, C.
H. Wiebusch, L. Wille, D. R. Williams, L. Wills, M. Wolf, J. Wood, T. R. Wood, K.
Woschnagg, G. Wrede, D. L. Xu, X. W. Xu, Y. Xu, J. P. Yanez, G. Yodh, S. Yoshida, T.
Yuan, M. Zöcklein, IceCube Collaboration, Search for Sources of Astrophysical
Neutrinos Using Seven Years of IceCube Cascade Events. Astrophys. J. 886, 12 (2019).
13. A. Albert et al., Joint Constraints on Galactic Diffuse Neutrino Emission from the
ANTARES and IceCube Neutrino Telescopes. Astrophys. J. Lett. 868, L20 (2018).
14. A. Albert et al., New constraints on all flavor Galactic diffuse neutrino emission with the
ANTARES telescope. Phys. Rev. D 96, 062001 (2017).
15. R. Abbasi et al., J. Instrum. 16, P07041 (2021).
16. The IceCube Collaboration, Combining maximum-likelihood with deep learning for event
reconstruction in IceCube, presented at 2021 ICRC The Astroparticle Physics
Conference, 12 to 23 Junly 2021, Berlin, Germany, 10.22323/1.395.1065 (2021).
17. M. Huennefeld, icecube/event-generator: Version 1.0.2. Zenodo (2022);
https://doi.org/10.5281/zenodo.7412035.
18. M. Aartsen et al., The IceCube Neutrino Observatory: Instrumentation and online systems. J.
Instrum. 12, P03012 (2017).
19. M. Aartsen et al., J. Instrum. 9, P03009 (2014).
20. M. G. Aartsen et al., Time-Integrated Neutrino Source Searches with 10 Years of IceCube
Data. Phys. Rev. Lett. 124, 051103 (2020).
21. R. Abbasi et al., IceCube high-energy starting event sample: Description and flux
characterization with 7.5 years of data. Phys. Rev. D 104, 022002 (2021).
22. M. G. Aartsen et al., Characteristics of the diffuse astrophysical electron and tau neutrino
flux with six years of IceCube high energy cascade data. Phys. Rev. Lett. 125, 121104
(2020).
23. T. K. Gaisser, M. Honda, Flux of atmospheric neutrinos. Annu. Rev. Nucl. Part. Sci. 52, 153–
199 (2002).
24. S. Schönert, T. K. Gaisser, E. Resconi, O. Schulz, Vetoing atmospheric neutrinos in a high
energy neutrino telescope. Phys. Rev. D Part. Fields Gravit. Cosmol. 79, 043009 (2009).
25. T. K. Gaisser, K. Jero, A. Karle, J. van Santen, Generalized self-veto probability for
atmospheric neutrinos. Phys. Rev. D Part. Fields Gravit. Cosmol. 90, 023009 (2014).
26. M. G. Aartsen et al., Atmospheric and astrophysical neutrinos above 1 TeV interacting in
IceCube. Phys. Rev. D Part. Fields Gravit. Cosmol. 91, 022001 (2015).
27. IceCube Collaboration, Evidence for high-energy extraterrestrial neutrinos at the IceCube
detector. Science 342, 1242856 (2013).
28. Y. LeCun et al., Advances in Neural Information Processing Systems 2 (Morgan-Kaufmann,
1990), pp. 396–404.
29. R. Abbasi et al., J. Instrum. 16, P08034 (2021).
30. Materials and methods are available as supplementary materials.
31. J. Braun, J. Dumm, F. De Palma, C. Finley, A. Karle, T. Montaruli, Methods for point source
analysis in high energy neutrino telescopes. Astropart. Phys. 29, 299–305 (2008).
32. M. Su, T. R. Slatyer, D. P. Finkbeiner, Giant gamma-ray bubbles from Fermi-LAT: Active
galactic nucleus activity or bipolar galactic wind? Astrophys. J. 724, 1044–1082 (2010).
33. D. Gaggero, D. Grasso, A. Marinelli, A. Urbano, M. Valli, The gamma-ray and neutrino sky:
A consistent picture of Fermi-LAT, Milagro, AND IceCube results. Astrophys. J. Lett.
815, L25 (2015).
34. S. Wakely, D. Horan, TeVCat, http://tevcat2.uchicago.edu (2022).
35. A. Voruganti, C. Deil, A. Donath, J. King in 6th International Symposium on High Energy
Gamma-Ray Astronomy (AIP Conference Proceedings, 2017) 1792, p. 070005;
http://gamma-sky.net/.
36. R. Abbasi et al., Time-integrated searches for point-like sources of neutrinos with the 40-
string IceCube detector. Astrophys. J. 732, 18 (2011).
37. M. Amenomori et al., First Detection of sub-PeV diffuse gamma rays from the Galactic disk:
Evidence for ubiquitous Galactic cosmic rays beyond PeV energies. Phys. Rev. Lett. 126,
141101 (2021).
38. R. Abbasi et al., Improved characterization of the astrophysical muon–neutrino flux with 9.5
years of IceCube data. Astrophys. J. 928, 50 (2022).
39. A. Mellinger, A color all-sky panorama image of the Milky Way. Publ. Astron. Soc. Pac.
121, 1180–1187 (2009).
40. Fermi’s 12-year view of the gamma-ray sky, NASA Goddard Space Flight Center (2022);
https://svs.gsfc.nasa.gov/14090.
41. T. Chen, C. Guestrin in Proceedings of the 22nd ACM SIGKDD International Conference on
Knowledge Discovery and Data Mining (Association for Computing Machinery, 2016),
p. 785.
42. M. G. Aartsen et al., All-sky search for time-integrated neutrino emission from astrophysical
sources with 7 yr of IceCube data. Astrophys. J. 835, 151 (2017).
43. A. Fedynitch, R. Engel, T. K. Gaisser, F. Riehn, T. Stanev, Calculation of conventional and
prompt lepton fluxes at very high energy. EPJ Web Conf. 99, 08001 (2015).
44. T. K. Gaisser, Spectrum of cosmic-ray nucleons, kaon production, and the atmospheric muon
charge ratio. Astropart. Phys. 35, 801–806 (2012).
45. T. K. Gaisser, T. Stanev, S. Tilav, Cosmic ray energy spectrum from measurements of air
showers. Front. Phys. 8, 748–758 (2013).
46. A. Fedynitch, F. Riehn, R. Engel, T. K. Gaisser, T. Stanev, Hadronic interaction model sibyll
2.3 c and inclusive lepton fluxes. Phys. Rev. D 100, 103018 (2019).
47. C. A. Argüelles, S. Palomares-Ruiz, A. Schneider, L. Wille, T. Yuan, Unified atmospheric
neutrino passing fractions for large-scale neutrino telescopes. J. Cosmol. Astropart. Phys.
2018, 047 (2018).
48. M. Huennefeld et al., in Proceedings of the 35th International Cosmic Ray Conference.
Proceedings of Science ICRC2017, 1057 (2018).
49. R. A. Fisher, Dispersion on a sphere. Proc. R. Soc. Math. Phys. Eng. Sci. 217, 295–305
(1953).
50. M. Aartsen et al., Efficient propagation of systematic uncertainties from calibration to
analysis with the SnowStorm method in IceCube. J. Cosmol. Astropart. Phys. 10, 048
(2019).
51. A. W. Strong, I. V. Moskalenko, Propagation of cosmic‐ray nucleons in the galaxy.
Astrophys. J. 509, 212–228 (1998).
52. C. Evoli, D. Gaggero, D. Grasso, L. Maccione, Cosmic ray nuclei, antiprotons and gamma
rays in the galaxy: A new diffusion model. J. Cosmol. Astropart. Phys. 2008, 018 (2008).
53. C. Evoli, D. Gaggero, D. Grasso, L. Maccione, Common solution to the cosmic ray
anisotropy and gradient problems. Phys. Rev. Lett. 108, 211102 (2012).
54. D. Gaggero, D. Grasso, A. Marinelli, A. Urbano, M. Valli, KRAγ cosmic-ray diffusion
models. Zenodo (2022); https://doi.org/10.5281/zenodo.7070823.
55. K. Fang, K. Murase, Multimessenger implications of sub-PeV diffuse Galactic gamma-ray
emission. Astrophys. J. 919, 93 (2021).
56. K. Murase, M. Ahlers, B. C. Lacki, Testing the hadronuclear origin of PeV neutrinos
observed with IceCube. Phys. Rev. D Part. Fields Gravit. Cosmol. 88, 121301 (2013).
57. M. Ahlers, K. Murase, Probing the Galactic origin of the IceCube excess with gamma rays.
Phys. Rev. D Part. Fields Gravit. Cosmol. 90, 023010 (2014).
58. M. Ahlers, F. Halzen, Opening a new window onto the universe with IceCube. Prog. Part.
Nucl. Phys. 102, 73–88 (2018).
59. G. J. Feldman, R. D. Cousins, Unified approach to the classical statistical analysis of small
signals. Phys. Rev. D Part. Fields 57, 3873–3889 (1998).
60. S. Abdollahi, F. Acero, M. Ackermann, M. Ajello, W. B. Atwood, M. Axelsson, L. Baldini,
J. Ballet, G. Barbiellini, D. Bastieri, J. Becerra Gonzalez, R. Bellazzini, A. Berretta, E.
Bissaldi, R. D. Blandford, E. D. Bloom, R. Bonino, E. Bottacini, T. J. Brandt, J. Bregeon,
P. Bruel, R. Buehler, T. H. Burnett, S. Buson, R. A. Cameron, R. Caputo, P. A. Caraveo,
J. M. Casandjian, D. Castro, E. Cavazzuti, E. Charles, S. Chaty, S. Chen, C. C. Cheung,
G. Chiaro, S. Ciprini, J. Cohen-Tanugi, L. R. Cominsky, J. Coronado-Blázquez, D.
Costantin, A. Cuoco, S. Cutini, F. D’Ammando, M. DeKlotz, P. Torre Luque, F. de
Palma, A. Desai, S. W. Digel, N. D. Lalla, M. D. Mauro, L. D. Venere, A. Domínguez,
D. Dumora, F. F. Dirirsa, S. J. Fegan, E. C. Ferrara, A. Franckowiak, Y. Fukazawa, S.
Funk, P. Fusco, F. Gargano, D. Gasparrini, N. Giglietto, P. Giommi, F. Giordano, M.
Giroletti, T. Glanzman, D. Green, I. A. Grenier, S. Griffin, M.-H. Grondin, J. E. Grove,
S. Guiriec, A. K. Harding, K. Hayashi, E. Hays, J. W. Hewitt, D. Horan, G. Jóhannesson,
T. J. Johnson, T. Kamae, M. Kerr, D. Kocevski, M. Kovac’evic’, M. Kuss, D. Landriu, S.
Larsson, L. Latronico, M. Lemoine-Goumard, J. Li, I. Liodakis, F. Longo, F. Loparco, B.
Lott, M. N. Lovellette, P. Lubrano, G. M. Madejski, S. Maldera, D. Malyshev, A.
Manfreda, E. J. Marchesini, L. Marcotulli, G. Martí-Devesa, P. Martin, F. Massaro, M.
N. Mazziotta, J. E. McEnery, I. Mereu, M. Meyer, P. F. Michelson, N. Mirabal, T.
Mizuno, M. E. Monzani, A. Morselli, I. V. Moskalenko, M. Negro, E. Nuss, R. Ojha, N.
Omodei, M. Orienti, E. Orlando, J. F. Ormes, M. Palatiello, V. S. Paliya, D. Paneque, Z.
Pei, H. Peña-Herazo, J. S. Perkins, M. Persic, M. Pesce-Rollins, V. Petrosian, L. Petrov,
F. Piron, H. Poon, T. A. Porter, G. Principe, S. Rainò, R. Rando, M. Razzano, S.
Razzaque, A. Reimer, O. Reimer, Q. Remy, T. Reposeur, R. W. Romani, P. M. S.
Parkinson, F. K. Schinzel, D. Serini, C. Sgrò, E. J. Siskind, D. A. Smith, G. Spandre, P.
Spinelli, A. W. Strong, D. J. Suson, H. Tajima, M. N. Takahashi, D. Tak, J. B. Thayer, D.
J. Thompson, L. Tibaldo, D. F. Torres, E. Torresi, J. Valverde, B. V. Klaveren, P. Zyl, K.
Wood, M. Yassine, G. Zaharijas, Fermi Large Area Telescope Fourth Source Catalog.
Astrophys. J. Suppl. Ser. 247, 33 (2020).
61. G. Illuminati et al., in Proceedings of the 37th International Cosmic Ray Conference.
Proceedings of Science ICRC2021, 1161 (2021).
62. J. Aublin et al., in Proceedings of the 36th International Cosmic Ray Conference.
Proceedings of Science ICRC2019, 840 (2019).