Draft version July 5, 2016

Preprint typeset using LATEX style emulateapj v. 5/2/11

HIGH-ENERGY NEUTRINOS FROM SOURCES IN CLUSTERS OF GALAXIES

Ke Fang1,2 , Angela V. Olinto

arXiv:1607.00380v1 [astro-ph.HE] 1 Jul 2016

Draft version July 5, 2016

ABSTRACT
High-energy cosmic rays can be accelerated in clusters of galaxies, by mega-parsec scale shocks
induced by accretion of gas during the formation of large-scale structure, or by powerful sources
harbored in clusters. Once accelerated, the highest energy particles leave the cluster via almost
rectilinear trajectories, while lower energy ones can be confined by the cluster magnetic field up
to cosmological time and interact with the intracluster gas. Using a realistic model of the baryon
distribution and the turbulent magnetic field in clusters, we studied the propagation and hadronic
interaction of high-energy protons in the intracluster medium. We report the cumulative cosmic ray
and neutrino spectra generated by galaxy clusters including embedded sources, and demonstrate that
clusters can contribute a significant fraction of the observed IceCube neutrinos above 30 TeV while
remaining undetected in high-energy cosmic rays and rays for reasonable choices of parameters and
source scenarios.
1. INTRODUCTION

The IceCube Observatory has observed the first highenergy astrophysical neutrinos (Aartsen et al. 2013a,b,c).
The all-flavor diffuse neutrino flux is reported to be =
2.06 1018 (E /105 GeV)2.46 GeV1 cm2 sr1 s1 for
the energy range 25 TeV < E < 1.4 PeV (Aartsen et al.
2015). Searches for small-scale anisotropies from the
three-year IceCube data do not see any significant clustering or correlations (Aartsen et al. 2015). The origin
of these high-energy neutrinos remains unclear (Murase
2015).
Accretion of gas during the formation of the largescale structure can give rise to mega-parsec scale shocks
that accelerate high-energy cosmic rays (Miniati et al.
2000; Ryu et al. 2003), and even ultra-high energy cosmic
rays (UHECRs) if the medium around the shocks contain
heavy nuclei (Inoue et al. 2007). Numerical simulations
(Vazza et al. 2014; Hong et al. 2014; Miniati 2015) also
suggest the possibility of stochastic particle acceleration
in large-scale structures, though the flux level of cosmic
rays depend on the Mach number and the location of
shocks (Vazza et al. 2014).
Powerful sources harbored in the galaxy clusters can
also accelerate particles to high energies. Many plausible candidate sources have been proposed in literature,
including steady sources like active galactic nuclei (AGN)
(e.g. Stecker et al. (1991); Winter (2013); Murase et al.
(2014); Dermer et al. (2014)), and transients like gammaray bursts (GRBs) (see Mesz
aros (2006) for review), fastspinning newborn pulsars (Blasi et al. 2000; Fang et al.
2012, 2013), magnetars (Arons 2003; Murase et al. 2009),
and blazar flares (Farrar & Gruzinov 2009).
Clusters are famously known as cosmic ray reservoirs,
due to their ability to confine cosmic rays with turbulent magnetic fields up to cosmological time (Volk et al.
(1996); Berezinsky et al. (1997), also see Brunetti &
1 University of Maryland, Department of Astronomy, 1105
PSC, College Park, MD 20742, USA
2 Joint Space-Science Institute, College Park, MD, 20742
3 Department of Astronomy & Astrophysics, Kavli Institute
for Cosmological Physics, The University of Chicago, Chicago,
Illinois 60637, USA

Jones (2015) for a recent review). Hence the accelerated cosmic rays have a good chance to interaction with
the intracluster medium (ICM), leading to the production of secondary neutrinos and rays. A cosmic ray
reservoir scenario is favored to explain the absence of detection of neutrinos above a few PeV, especially around
the Glashow resonance at 6.3 PeV. This is because rather
than otherwise a mysterious stop in the cosmic ray spectrum, a cosmic ray reservoir scenario naturally introduces a spectral softening caused by the faster escape
of the higher-energy cosmic rays in a magnetized source
environment.
Many analytical and semi-analytical works were conducted to calculate these secondary fluxes (Dar & Shaviv 1995; Volk et al. 1996; Berezinsky et al. 1997; de
Marco et al. 2006; Murase et al. 2008; Wolfe et al. 2008;
Murase et al. 2013). However most of the analytical approaches adopted an overly simplified modeling of the
ICM by assuming both uniform gas distribution and
uniform magnetic field. Colafrancesco & Blasi (1998)
took into account the density profile and mass function
of galaxy clusters, yet still assumed a uniform magnetic
field within each cluster and the same cosmic ray diffusion length for all groups of clusters with different masses.
Numerical propagation of cosmic rays in more realistic
three-dimensional cluster magnetic fields were explored
by Rordorf et al. (2004) and Kotera et al. (2009), but
the studies were limited to neutrinos from a single type
of clusters.
Numerical modeling of the non-uniform gas distribution and magnetic field structure is crucial in an environment like the ICM, because both duration and location
of the confinement of the charged particles impact the interaction rate, and thus an accurate computation of the
neutrino production. Moreover, although massive clusters could be brighter cosmic ray and neutrino sources,
they are far less common than medium-sized clusters. It
is thus unclear how the mass and redshift dependence
of cluster number density would impact the total neutrino spectrum. It is also unknown which group of clusters would dominate PeV cosmic rays and thus be more
relevant for the detected neutrinos. The latter point is

2
also important for revealing the possibility of pinpointing sources with the increasing statistics of IceCube as
well as future experiments (Ahlers & Halzen 2014; Fang
& Miller 2016).
Bechtol et al. (2015) pointed out that new studies of
the blazar flux at gamma rays above 50 GeV result in a
lower residual non-blazar component of the extragalactic gamma-ray background. This puts a tight constraint
on neutrino sources that are transparent to gamma rays
(Murase et al. 2016), especially those which produce
neutrinos through hadronuclear (pp) interactions (as in
galaxy clusters (Kotera et al. 2009)). However, this constraint is drawn based on the assumption that the pp
scenario has a E 2 spectrum extended below 10 TeV,
which is not necessarily valid for all source types. For
example, particles accelerated by fast-spinning newborn
pulsars (Blasi et al. 2000; Fang et al. 2012, 2013) can have
a spectrum index less than 2, and cosmic rays accelerated
in the AGN (Murase et al. 2012; Dermer et al. 2012) can
present a cutoff at low energies due to the confinement
of the source environment.
In this paper, we investigate high-energy cosmic rays
and neutrinos from clusters, by numerically propagating
cosmic rays down to TeV in galaxy clusters with a wide
range of masses and redshifts. We limit the uncertainties
of our results with a more realistic modeling of the ICM
gas and the turbulent cluster magnetic field than some of
the previous works, and by adopting the mass accretion
rates, cluster baryon fraction, and halo mass function as
constrained by cosmological observations.
We report the integrated cosmic ray and neutrino spectra from the entire cluster population in two scenarios:
1. the accretion shock scenario: cosmic rays are accelerated by the cluster accretion shocks and injected
at the outskirts of clusters;
2. the central source scenario: cosmic rays are accelerated by sources at the centers of the clusters.
Our results demonstrate that neutrinos from cluster accretion shocks could contribute <
20% of the IceCube
flux; however, if bright astrophysical sources inside the
galaxy clusters can accelerate particles with an injection
spectrum index below 2, clusters could reproduce both
the spectrum and flux of IceCube neutrinos above 30
TeV, while remaining consistent with the measurements
of high-energy cosmic rays and rays.
2. MODELS

In this section we lay out the models used in our work.

We first review the particle acceleration in clusters in
Sec. 2.1, then examine the diffusion and confinement of
cosmic rays in the turbulent cluster magnetic field in
Sec. 2.2. We calculate the neutrino production in a single cluster in Sec. 2.3, and discuss the formalism used to
integrate the source contributions in Sec. 2.4.
2.1. Cosmic Ray Acceleration

Charged particles can get accelerated by large-scale

shocks induced by cluster accretion activities, or (and)
produced by the powerful sources harbored inside the
clusters. Below we discuss the two scenarios accordingly.
The spatial scale of the accretion shocks can be estimated by the virial radius of the host cluster. For a

cluster with mass M = M15 1015 M at redshift z, the

size of the accretion shock is

1/3
3M
rsh rvir =
(1)
4 vir (z)m (0)

1/3
vir (0)
1/3
= 2.6 M15
Mpc
vir (z)
where 4vir (z) (18 2 + 82x 39x2 )/(x + 1) with
x = M (z) 1 is the virial overdensity with respect
to the mean matter density (Hu & Kravtsov 2003). For
simplicity, the cosmological term vir (0)/vir (z) will be
ignored in the analytical estimations below (for reference,
vir (0)/vir (1) 1.6).
The typical velocity of accretion shocks should be
comparable to the free-fall velocity (Inouepet al. 2007;
Murase et al. 2008), which reads vsh
GM /rsh =
1/3
1
1300 M15 km s . The acceleration time in a strong
2
shock can be estimated by tacc = 20 Dsh /vsh
(Kang
et al. 1997; Inoue et al. 2007)), assuming a quasi-parallel
shock. Note this time can be much shorter for oblique
shocks with more general magnetic field angles (Jokipii
1987). Strong accretion shocks are expected at the
outskirts of clusters (Vazza et al. 2014). Taking the
Bohm limit, the diffusion coefficient of the cluster shock
reads Dsh = rL c/3, where rL = E/ZeB is the Larmor radius of a cosmic ray particle with energy E and
charge Z in a magnetic field B. In a shock with field
strength B = 1 B6 G, the acceleration time for a
particle to reach Ecr = 1018 Ecr,18 eV would be tacc =
2/3 1 1
1.3 Ecr,18 M15 B6
Z Gyr. On the other hand, the
maximum time cosmic rays stay in the shocks due to
2
diffusive escape is tesc rsh
/6 Dsh (Inoue et al. 2007).
tacc = tesc thus sets the maximum energy that cosmic
rays can be accelerated to:
2/3

Ecr,max = 2.8 1018 M15 Z B6 eV.

(2)

The energy budget of high-energy cosmic rays is determined by the accretion rate of the cluster. The mass accretion rate of halos can be described by a fitting function
hM i = 42 (M/1012 M )1.127 (1 + 1.17z)
E(z) M yr1
p
(McBride et al. 2009), with E(z) = m (1 + z)3 + .
Note that in this work we take m = 0.308, h = 67.8
(Planck Collaboration et al. 2015) and assume a flat Universe.
The kinetic energy of the accretion shocks is Lacc =
fb GM M /rsh , with fb = 0.13 (M/1014 M )0.16 being
the average baryon fraction of galaxy clusters (Gonzalez
et al. 2013). If a fraction fcr = 1% fcr,2 of this energy
is converted into cosmic rays, the luminosity of cosmic
rays from accretion shocks is then
1.95
Lcr = 2.0 1044 M15
fcr,2 erg s1

(3)

Notice that the cosmic ray luminosity of a cluster is

highly dependent on the cluster mass.
Particles accelerated in high-energy sources inside the
clusters can have very different spectra and energetics
depending on the source types. In general, E 2 spectrum would be expected from diffusive shock accelerations, but harder spectra are possible when the particle energy is converted from the electromagnetic energy

3
of the sources, for example, in pulsar magnetospheres
(Arons 2003), or in reconnection processes (Guo et al.
2014). Besides, an interplay between the acceleration
and dissipation processes at the acceleration site could
also lead to a hard injection spectrum (Murase et al.
2012; Dermer et al. 2012). Murase et al. (2016); Bechtol
et al. (2015) demonstrated that a pp scenario with a E 2
spectrum normalized to the IceCube datapoint at 10
TeV would overproduce rays above the flux of the nonblazar component of the isotropic -ray background. In
this work we will instead consider the case of an injection
spectrum index equal or smaller than 2 but normalized
to the IceCube datapoint at 30 TeV (which is about 3
times lower than that at 10 TeV). Specifically, in Sec. 3
we focus on a benchmark case of an injection spectrum
dN/dE E 1.5 and maximum energy Emax = 50 PeV.
We discuss the effects of different choices of dN/dE and
Emax in Sec. 4.
2.2. Cosmic Ray Diffusion

Galaxy clusters are known as efficient containers of cosmic rays (V

olk et al. 1996; Berezinsky et al. 1997). Depending on the particles Larmor radius rL and the fields
coherence length lc , the propagation of cosmic rays in
the cluster turbulent magnetic field can be divided into
semi-diffusive (rL > lc ) and diffusive (rL  lc ) regimes
(Kotera & Lemoine 2008).
When a particles energy reaches Esd = 9.3
1019 Z B6 lc,1 eV, the particle enters the semi-diffusive
regime with a diffusion coefficient
1 2 1
cr l
(4)
3 L c
However, most particles have energy much lower than
Esd , and thus stay in the diffusive regime. The diffusion
coefficient in this regime can be written as (Brunetti &
Jones 2014)

2
B
1
2w w1
c rL
lc
(5)
Dcl =
3 B

2/3 
1/3
lc
E
6.8 1029
0.1 Mpc
1GeV
Dcl =

(B6 Z)1/3 cm2 s1

where w = 5/3 is the spectrum index for Kolmogorov
diffusion, B B and l0 = l0,1 0.1 Mpc are the typical amplitude and scale of magnetic field fluctuations in
massive clusters (Brunetti & Jones 2015).
The diffusion time of cosmic rays in the cluster magnetic field can be estimated as
tdiff

2
rvir
2/3 1/3
1/3 2/3
= 1.5 M15 E18 Z 1/3 B6 lc,1 Gyr (6)
2Dcl

Cosmic rays with diffusion time longer than the Hubble

time tH 14 Gyr would be completely confined by the
cluster magnetic field, defining the lower limit of the energy of cosmic rays we expect to observe from the cluster
as
2
2
Ecr,conf = 1.4 1015 M15
Z l0,1
B6 eV
(7)
Particles with E > Ecr,conf would have a better chance
to escape from the cluster magnetic field. To estimate

this escape probability, we first write down the probability that a particle with diffusion coefficient D reaches
radius r after time t:
33/2 3r2 /4Dt
e
(8)
4Dt
assuming that the particle starts from r = 0 and the
field is homogenous. The probability that a particle can
successfully escape from a cluster of size rvir within the
Hubble time is then (Kotera & Lemoine 2008)
Z
fesc
dr p(r, tH ) 4r2
(9)
p(r, t) =

rvir

Cosmic rays accelerated in the strong shocks in the

accretion region are expected to have a spectrum index
2. The flux of high-energy cosmic rays from a cluster
with mass M = 1015 M at redshift z = 0.1 can be
calculated as
1
2
2 dN
CR = ECR
ECR
fesc (1 f )
(10)
dE 4d2L

2 
2
E
dL
= 0.12
1 PeV
dL (z = 0.1)
1.95
M15
fCR,2 fesc (1 f ) eV cm2 s1 sr1

where dL is the luminosity distance, and f is the pion

production rate that will be introduced in Sec. 2.3. Note
that we have taken 2.1 in the above estimation. In
case of a much softer spectrum, an additional correction
term ( 2)(PeV/GeV)2 originated from spectrum
normalization should be included.
2.3. Neutrino Production
Accelerated cosmic rays would propagate in the cluster
magnetic filed and interact with both hadron and photon
ambience of the ICM. The photon ambience of the ICM
includes the Cosmic Microwave Background (CMB) and
the intracluster infrared background, but both are subdominant when compared to the hadronic background
as shown in Kotera et al. (2009). We thus focus on the
hadron background in this work.
The density profile of the thermal gas in the ICM can
be described by the classical model (Cavaliere &
Fusco-Femiano 1976):
"
 2 #3/2
r
nICM (r) = nICM,0 1 +
(11)
rc

where = 0.8 and rc 0.1 rvir is the core radius (Fujita

et al. 2003). The density profile is normalized by
Z
fb M = ICM (r)4r2 dr
(12)
The mass density is related to the number density
through ICM = nICM mp , with 0.61 being the
mean molecular weight.
The magnetic field distribution should trace the baryon
distribution in the cluster through the flux conservation
2/3
B nth (Brunetti et al. 2004). Specifically, in clustersized halos, which we define here to consist of halos
greater than 1012 M , the magnetic field as a function of

x / Mp

1
2

Mp

y/

x / Mp

Mp

1
2

y/

1
2

z / Mpc

z / Mpc

Fig. 1. Trajectory of high-energy cosmic rays in the turbulent magnetic field of a model cluster with mass M = 1015 M located at
redshift z = 0.1. The grey sphere represents a cluster. The virial radius of the simulated case is 2.6 Mpc just outside the grey sphere. The
baryon and the magnetic field distribution in the ICM follows the model described by equation 11 and 13. The central magnetic field
strength is B0 = 10 G. The left panel shows a proton with injection energy Ep = 1019 eV which has a Larmor radius comparable to the
coherence length of the B field. The right panel shows the trajectories of a proton with injection energy Ep = 1017 eV and its interaction
products. The starting point of the proton is at the center of the cluster. The trajectories of the primary proton (blue), the secondary
protons (green), as well as the neutron and neutrino products (red) are indicated by a series of points at each propagation step.

the radial distance from the center of the cluster is given

by

 "
 2 #
M
r
B(M, r) = B0
1+
(13)
M0
rc
where B0 is the central magnetic field at a mass M0
= 1014 M , and controls the dependence of B0 on
the mass of the cluster in consideration. In our default
model, we set B0 = 3 G and = 0.0 (V
olk & Atoyan
1999; Beck 2012). Following the distribution of nICM ,
we define the core radius of the magnetic field model
to be equivalent to rc = 0.1 rvir and the falloff index
= 0.8. The magnetic fields in galaxy-sized halos are
instead dominated by the dense baryonic cores, and thus
have a different structure. We thus use a two part magnetic field model which adopts the larger of the two following magnetic fields as in Fang & Linden (2015):

This corresponds to a pion production rate:

0.5
fpp = hyICM ipp 6 103 M15
26

(16)

where pp 510
cm is the effective proton-proton
interaction cross section. Note that particles injected at
the cluster center would have a much greater chance to
be confined by the cluster magnetic fields which leads
to larger fpp . In contrast, particles injected near the
boundary may not even meet an ICM baryon before escaping from the cluster. Since the distribution of baryons
in the cluster is not uniform, and different confinement
time of particles in the cluster magnetic field would lead
to different fpp , in simulations this rate is more precisely
calculated by tracking particle propagation in the cluster
magnetic field (see Sec 3.1).

(14)

2.4. Integrated Neutrino Flux

We expect a diffusive neutrino flux integrated from the
entire clusters population. The differential number density, dn/dM , of clusters with mass M at redshift z is
given by the halo mass function:

where B1 = 7.6 G and R1 = 0.025 Rvir and B2 = 35 G

with R2 = 0.008Rvir . However we note that due to the
sharp cutoff around R1 and R2 , the magnetic field at
large radii, where accretion shocks most likely occur, are
too low to accelerate to high energies. The contribution
from shocks in low mass halos is rather negligible in our
scenario.
For one crossing of the cluster, a particle would expect
to confront a column density of
Z rvir
0.5
hyICM i
nICM (r) dr = 1.2 1022 M15
cm2 (15)

m d ln 1
dn
(M, z) = f ()
(17)
dM
M dM
where m is the mean density of the universe at the epoch
of analysis, m (z) = m (0) (1 + z)3 , and (M, z) is the
rms variance of the linear density field smoothed on a
top-hat window
R = (3M/4m )1/3 . SpecifR function
2
lin

(kR) is the
ically, = dk P (k)W (kR) k 2 , where W
Fourier transform of the real-space top-hat window function of radius R (Tinker et al. 2008). For f () we adopt
the mass function multiplicity described by Sheth & Tormen (1999).



B(r) = max B1 er/R1 , B2 er/R2

5
The integrated neutrino flux can be calculated as
Z
dn (1 + z)2 L (M, z) dV
2
E (E ) = dM
(18)
dM
4d2L
d
Z zmax
c dz
1
=
4 zmin H0 E(z)
Z Mmax
dn 2 dN

dM
E
((1 + z)E , M, z)
dM dE
Mmin
where dV = c dz/(H0 E(z))(1 + z)2 d2A d is the comoving
volume and dA = dL (1 + z)2 is the angular diameter
distance. The lower and upper limits of the integration
are taken to be Mmin = 1012 M and Mmax = 1016 M ,
since galaxies with M < Mmin barely contribute to the
energy window we are interested in, while clusters with
M > Mmax have too low a number density to significantly contribute to the total flux. The redshift integration goes from zmin = 0.01, where the closest galaxy
clusters are located, up to zmax = 5. As we will demonstrate in Sec. 3.3, the flux from the most distant clusters
is dominated by that from the closer ones, so our results are not sensitive to the choice of zmax as long as
zmax  1.
3. RESULTS
3.1. Numerical Setup
The interactions between high-energy proton and ICM
baryons were simulated by Monte Carlo as in Fang et al.
(2012). The cross sections and products of pp interactions were calculated based on the hadronic interaction
model EPOS (Werner et al. 2006). In addition, we implemented the public UHECR propagation code CRPropa3
(Kampert et al. 2013) to study the semi-diffusive propagation of high-energy cosmic rays in cluster magnetic
fields.
In our simulation, the baryon distribution in the ICM
is set to follow equation 11. The field is generated to
follow the Kolmogorov-type turbulence with spectrum
index w = 5/3 and have random directions with a coherence length lc 0.03 rvir (Brunetti & Jones 2015). The
strength of the magnetic field of the ICM is normalized
by equations 13 and 14. (We will discuss the impact of a
stronger or weaker B on our results in Sec. 4.) Therefore
in our model the cluster environment is fully determined
by two parameters: the cluster mass M and the redshift
z.
In the accretion shock scenario, we assume that a stationary shock is located at dsh = sh Rta (Colafrancesco
& Blasi 1998), where sh 0.347, and Rta 2 Rvir is
the turn-around radius. The maximum energy of the
injected cosmic rays is calculated by equation 2 taking
B = B(dsh ). Note that adopting the magnetic field
strength at the cluster boundary is a conservative assumption as the shocked region in a cluster is not limited to the outer skirts. The injection of cosmic rays is
assumed to follow a spectrum dN/dE E 2 normalized
by equation 3. The energy fraction fcr is left as a free
parameter to be determined by observation.
In the central source scenario, cosmic rays are injected
in isotropic directions from the center of the host cluster, with a universal maximum energy Emax = 50 PeV
and spectrum dN/dE E 1.5 . The effect of different

choices of Emax and dN/dE to our results will be discussed in Sec. 4. Considering that the total luminosity
of high-energy sources inside the cluster should scale to
the kinetic power of the cluster, we still refer to equation 3 as the energy budget of injected cosmic rays.
Finally, the diffusive propagation of each particle is
tracked until either it successfully leaves the cluster, or its
total propagation time exceeds the Hubble time. In the
accretion shock scenario, particles are injected isotropically from dsh 0.7 Rvir . Both magnetic field and number density of the ICM are low at the outskirt region,
hence some fraction of the particles can leave the cluster
within a relatively short time and nearly zero interaction,
whereas the others will be trapped toward the inner side
of the cluster. In the central source scenario, particles injected from the cluster center confront the densest region
of the ICM, and could be easily confined over the Hubble
time. Meanwhile secondary and higher-order protons resulted from the interactions can lead to further productions of neutrinos (an example of the presence of such
secondary particles are shown by the green points in the
right panel of Fig 1). The difference in the injection locations lead to quite different behaviors of particles in the
two scenarios. The 3-d propagation and particle tracking in our simulation are needed to describe the transportation and interaction of particles of all orders in the
non-uniform environment, especially when the system is
asymmetric as in the accretion shock scenario.
3.2. Cosmic Ray Trajectories
In this section we demonstrate the propagation of cosmic rays in the turbulent magnetic field of a model cluster
with mass M = 1015 M at redshift z = 0.1. The central
magnetic field strength is set to be B0 = 10 G. The field
strength at the outer skirts is then Bvir 0.6 G. The
number density of the ICM gas at the cluster center is
n0 = 1.1 102 cm3 , as calculated by equation 12.
In the left panel of Figure 1, we show the trajectory of a
proton with injection energy Ep = 1019 eV. Note that the
cluster radius is 2.7 Mpc although it looks smaller due to
the projection of the 3D sphere in the plots. The Larmor
1
radius of the particle is rL = 0.01 E19 B6
Mpc, comparable to the coherence length of the field which is set to
be lc = 0.07 Mpc. To study such a semi-diffusive propagation, we utilize the Cash-Karp propagation method of
CRPropa to track the particles entire trajectory. Each
step of the propagation is indicated as a blue point in
the plot, and the step size is determined dominantly by
the Larmor radius, because the mean free path of the
charged particle is much larger than rL . As expected,
the particle travels almost rectilinearly. In this realization, the particle escaped the cluster without interaction
with the ICM gas, and had a total trajectory length of
46 Mpc.
In the right panel of Figure 1, we show the trajectory
of a proton with injection energy Ep = 1017 eV. The Larmor radius of the particle is now 700 times less than
the coherence length of the field, placing it in the fully
diffusive regime. The computing cost for tracking all the
steps of the particle would be enormous. Instead, we use
the approximation algorithm demonstrated in Kotera &
Lemoine (2008). Specifically, taking advantage of the
fact that a particle would lose its initial direction after
traveling for a distance much longer than the Larmor

total
M <1014 M
1014 M <M <1015 M
M >1015 M
Auger ICRC2015
TA SD6yr
TALE 2014
KASCADE

16

17

18

19

logE [eV]

20

10-6
E2 [GeV cm2 s1 sr1 ]

106
105
104
103
102
101
100
10-1
10-215

Fig. 2. Cumulative cosmic ray flux from the accretion shocks of

galaxy clusters in comparison with observational data from KASCADE (Apel et al. 2013), Auger (Aab et al. 2015) and Telescope
Array (Fukushima 2015). Cosmic ray protons are injected with
a spectrum dN/dE E 2 and a fcr = 2% conversion rate from
accretion energy to cosmic rays. The solid line indicates the total
flux of cosmic rays that escape the cluster magnetic field within
the Hubble time. The contribution is decomposed into three mass
groups: M < 1014 M (dash dotted), 1014 M < M < 1015 M
(dashed), M > 1015 M (dotted). The central magnetic field
strength is assumed to be 3 G for all cluster halos in this scenario.

radius, the particle can be simulated as undergoing a

random walk with step size lc , but has each numerical
step correspond to an actual trajectory length lc2 /2D,
with D the diffusion coefficient at the particles current
location. The points in the right panel of Figure 1 hence
describe such a random walk with a uniform step size
lc . Note however these steps represent different actual
trajectory lengths due to the distinctive D at different
parts of the cluster. The blue points correspond to the
trajectory of the primary proton, the green ones indicate
that of the secondary protons stemming from the interaction with the ICM, and lastly the red ones mark the
neutron and neutrino products, including both electron
and muon neutrinos. The primary proton was never able
to leave the cluster, though the neutrons and some of the
secondary protons succeeded. We conclude this section
by noting that the trajectories are random and could be
very different from realization to realization.
3.3. Cosmic Ray and Neutrino Spectra
The integrated cosmic ray flux from the cluster population in the accretion shock scenario is shown in Figure 2.
In all clusters in this scenario, cosmic rays are injected as
protons following a power-law spectrum dN/dE E 2 .
The conversion rate from the kinetic energy of the cluster shocks to the cosmic rays is set to be fcr = 2% to
fit the observed neutrino flux. The turbulent magnetic
fields in clusters have a central strength B0 = 3 G, and
the baryon density of the ICM is scaled to the cluster
mass by n0 fb M 0.16 . The solid line indicates the
flux of the cosmic rays that could successfully leave the
clusters and be observed at the earth. The overall spectrum follows E 2 as the injection spectrum and is cut
off around 1017.5 eV. The contribution to cosmic ray flux
from clusters is subdominant in all energies comparing
to observation. The integrated cosmic ray flux in the
central source scenario has a similar behavior as in the

Accretion Shock Scenario

10-7
10-8

total
1012 M <M <1014 M
1014 M <M <1015 M
M >1015 M
IceCube 2015

10-9
10-10

0.01 <z <0.3

0.3 <z <1
z >1

10-1112 13 14 15 16 17 18 19 20

21

logE [eV]

10-6
E2 [GeV cm2 s1 sr1 ]

E2 cr [eV cm2 s1 sr1 ]

Central Source Scenario

10-7
10-8

total
1012 M <M <1014 M
1014 M <M <1015 M
M >1015 M
IceCube 2015

10-9
10-10

0.01 <z <0.3

0.3 <z <1
z >1

10-1112 13 14 15 16 17 18 19 20
logE [eV]

Fig. 3. Cumulative neutrino flux from the galaxy clusters

compared to the IceCube observations (Aartsen et al. 2015). Top:
accretion shock scenario: as in Figure 2, the cosmic ray injection
follows dN/dE E 2 , fcr = 2%, and a maximum energy following
equation 2, determined by particle acceleration and escape in the
accretion shocks. Bottom: central source scenario: the cosmic ray
injection follows dN/dE E 1.5 , fcr = 0.5%, and Emax = 5
1016 eV. The blue thin lines decompose the contribution into three
mass groups: M < 1014 M (dash dotted) and 1014 M < M <
1015 M (dashed) and M > 1015 M (dotted). The green thin
lines instead decompose the flux into three redshift bins: 0.01 <
z < 0.3 (dashed), 0.3 < z < 1 (solid) and z > 1(dotted).

accretion shock scenario, except that we assume the flux

follows a E 1.5 injection spectrum.
We further decompose the flux into the contribution
from three mass groups: M < 1014 M , 1014 M <
M < 1015 M , M > 1015 M , shown in dash dotted,
dashed, and dotted lines correspondingly. As the cosmic
ray luminosity significantly depends on the cluster mass
through Lcr M 1.95 , clusters below 1014 M are not
as luminous as the massive clusters. In addition, since
Emax M 2/3 , particles from these clusters cannot reach
the highest energies. However, since the strength and
scale of magnetic fields in these clusters are also smaller,
particles have a better chance to leave the source. The
combined impact of these factors leads to a contribution of this mass group to 1/3 of the total cosmic
ray flux around PeV, and 10% above 1017 eV. The
1014 M < M < 1015 M group is found to make the
largest contribution to the flux up to the ankle around
1018 eV.
Above the ankle, clusters with mass M > 1015 M

7
are needed to produce UHE protons. However, due to
the exponential cutoff in the halo mass function for such
extremely massive clusters, a significant production of
UHECRs from clusters is not found given a pure proton
injection. The main contribution of this proton component is relevant to explain the second knee observed
around 1017 eV (Apel et al. 2013). Heavier nuclei, if injected at the shocks, may reach ultra-high energies, but
their neutrino production would be less significant.
As displayed in Figure 1, within the same cluster, cosmic rays with lower energy have a lower chance to leave
the system due to their smaller Larmor radius. Consequently, the spectrum of the escaped cosmic rays is
slightly harder than that at injection.
The cumulative neutrino flux from clusters is shown
in Figure 3. The top panel corresponds to the accretion
shock scenario. The injection of cosmic rays and the
cluster environment are set up as in Figure 2. The total
neutrino flux is indicated by the solid blue line. With
fcr = 2 %, clusters contribute a neutrino flux less than
20% of the IceCube detections. Unlike the cosmic ray
spectrum which mainly scales with the escape probability
fesc , the neutrino spectrum is rather sensitive to the pion
production rate f . The spectral index is found to be
about 2 around TeV, 2.4 at PeV and cut off above 10
PeV. The spectrum softening at PeV is a combined effect
of diffusion (lower f at higher energies) and cluster mass
function (higher flux but lower rate for massive clusters).
The bottom panel of Figure 3 shows the integrated neutrino flux in the central source scenario. The spectrum
below 100 TeV is mainly determined by the injection
and thus follows E 1.5 . Interestingly, at higher energies the softening effect due to the shorter cosmic ray
diffusion time and the lower population of massive clusters dominates over the hard injection. As a result, with
fcr = 0.5% this scenario could reproduce the spectrum
and flux of the IceCube measurement above 30 TeV.
As in Figure 2, we also decompose the neutrino contribution into three mass groups: M < 1014 M (dash
dotted), 1014 M < M < 1015 M (dashed) and M >
1015 M (dotted), all indicated by blue lines. The significance of the three mass groups contribution to the
neutrino fluxes follows a similar order as that of cosmic
rays, except that without the fesc factor impacting the
neutrino production, the M < 1014 M group now contributes less than 5 % of the total flux. The thin green
lines in Figure 2 decompose the contribution into three
redshift bins: 0.01 < z < 0.3 (dashed), 0.3 < z < 1
(solid) and z > 1(dotted). We find that due to the intense source distribution in the region 0.3 < z < 1, this
group makes the largest contribution to neutrino production. Due to the distance and the rare rate of massive
clusters at high redshift, clusters at z > 1 barely contribute.
4. DISCUSSION AND CONCLUSION

Unlike other astrophysical sources, galaxy clusters offer a unique environment for TeV-PeV neutrino production through efficient acceleration and confinement of
high-energy cosmic rays. By propagating particles in
three-dimensional turbulent magnetic fields and recording their interactions with the ICM gas, we find that
the integrated neutrino flux from the cluster accretion
shocks could account for 20% of the IceCube detec-

tions, while neutrinos produced by the interaction of cosmic rays from powerful sources hosted by clusters could
explain both the flux and spectrum of the IceCube data
above 30 TeV, if the injection spectrum is harder than
2. The lack of neutrino clustering around known sources
also fits well with the cluster production model. In addition, the high-energy cosmic rays that succeed in escaping from the clusters contribute less than 20% of the
observed cosmic ray flux around and below the ankle,
and may explain the second knee feature when galactic
and ultrahigh energy cosmic ray accelerators contributions are added to the clusters contribution.
Our results demonstrate that when taking into account
the effect of cosmic ray confinement and the mass dependence of the cluster number density, the integrated neutrino spectrum would conserve the injection spectrum
around TeV - 100 TeV, but become much steeper above
PeV. If high-energy sources harbored in the galaxy clusters could produce cosmic rays with spectrum harder
than E 2 , the tension between the neutrino-associated
GeV rays and the Fermi measurement of isotropic diffusive -ray background (Ackermann et al. 2015; Murase
et al. 2016) can be alliviated. Also notice that in the
cluster scenario, the dominant contribution comes from
sources beyond z 0.3 (as suggested by Fig. 3). Thus,
rays with E >
0.1 TeV are attenuated by the time they
arrive on Earth, through interactions with photons of
the extragalactic background light (EBL) and the CMB
that have a non-trivial optical depth, (z 0.3) > 1
(Stecker et al. 2006).
The search for the first rays from galaxy clusters
is still ongoing. No significant spatially extended ray emission from the nearby galaxy clusters was found
in four years of Fermi-LAT data, establishing limits
on the cosmic ray to thermal pressure ratio, XCR , to
be below 1.4% (Ackermann et al. 2014). The parameter fCR of our model can be translated to XCR by
XCR = PCR /Pth 0.6% fCR,2 (M tcr,conf /M ), where
PCR (1/3) fCR (GM ICM tcr,conf /rvir ) is the cosmic ray
pressure after an accumulation of particles for tcr,conf ,
and Pth = nICM (GM mp /2 rvir ) is the thermal pressure.
The fCR < 2% suggested by our model is thus consistent with the constraint from -ray searches. However,
the Fermi limit would constrain cluster scenarios with
fCR > 2%, corresponding to an injection index > 2.1.
In the central source scenario (bottom panel of Fig. 3)
we showed a benchmark case with an injection spectrum index = 1.5 and maximum cosmic ray energy
Emax = 50 PeV. The results are solid under moderate
adjustments on either or Emax (including adding a
mass dependence). However, cannot be as large as
2, otherwise the -ray counterpart will clearly exceed the
Fermi measurements of the isotropic -ray background.
Furthermore, if Emax  50 PeV, the injected cosmic rays
wouldnt be energetic enough to produce the PeV neutrinos. On the other hand, if Emax  50 PeV in all clusters,
the resulted neutrino spectrum could overshoot the null
bins at 5 - 10 PeV.
In our simulations we assumed a central magnetic field
strength of 3 G. In case of a smaller central magnetic
field strength B0 , highest energy cosmic rays would have
a higher chance to leave the system due to the weaker
magnetic field and less interaction materials in the en-

8
vironment. Lower energy cosmic rays would be less impacted, since they are confined to a relatively small volume. Hence the overall neutrino spectrum would be expected to have a lower flux but a softer spectrum. Conversely, stronger B0 would lead to a higher neutrino flux
due to the intenser confinement.
Neither of our scenarios could account for the IceCube
data below 30 TeV. Although majority of the neutrinos
are expected to come from extragalactic sources, Galactic sources could potentially contribute some fraction of
the flux (Ahlers et al. 2016), especially below 200 TeV
(Murase 2015). It is also possible that other types of
sources could contribute to this energy range.
Our conclusion on the accretion shock scenario is consistent with that from Zandanel et al. (2015), though
the two works have distinctive approaches. While Zandanel et al. (2015) derived the neutrino luminosity from
a scaling between radio and gamma-ray luminosities, we
directly simulated the particle propagation to obtain the
neutrino spectrum. Zandanel et al. (2015) did not assume any cosmic ray spectral steepening due to the escape of high-energy cosmic rays, whereas in our work this
cutoff was shown as a natural result from the energydependent transportation, and crucial for the survival of
the central source scenario. On the other hand, Zandanel
et al. (2014, 2015) indicated that the neutrino contribution from clusters could be further limited due to the fact

that not all clusters are expected to produce hadronic

emissions. The same constraint could also apply to our
results in the accretion shock scenario.
Although consistent with current -ray and radio detection limits, the cluster scenario could be robustly
tested by the growing statistics of IceCube as well as
-ray searches of galaxy clusters in the near future. If
the cluster contribution to the diffusive -ray background
and the cosmic ray to thermal pressure ratio can be further constrained by the -ray observations, a significant
contribution to IceCube neutrinos from galaxy clusters
would be ruled out. On the other hand, as our model
predicts a dominant neutrino flux from the most massive clusters above 5 PeV, future detection of strong
anisotropy in this energy regime would provide a firm
support to the cluster scenario.
We thank Cole Miller, Kohta Murase for helpful discussions. KF acknowledges the support of a Joint SpaceScience Institute prize postdoctoral fellowship. AO acknowledge financial support from the NSF grant NSF
PHY-1412261 and the NASA grant 11-APRA-0066 at
the University of Chicago, and the grant NSF PHY1125897 at the Kavli Institute for Cosmological Physics.
This work made use of computing resources and support
provided by the Research Computing Center at the University of Chicago.

REFERENCES
Aab, A. et al. 2015, Proceedings of the 34th International Cosmic
Ray Conference (ICRC 2015)
Aartsen, M. et al. 2013a, Science, 342, 1242856
. 2013b, Phys.Rev.Lett., 111, 021103
. 2013c, Astrophys.J., 779, 132
Aartsen, M. G., Ackermann, M., Adams, J., Aguilar, J. A.,
Ahlers, M., Ahrens, M., Altmann, D., Anderson, T., Arguelles,
C., Arlen, T. C., & et al. 2015, PRD, 91, 022001
Aartsen, M. G. et al. 2015, Astropart. Phys., 66, 39
Ackermann, M., Ajello, M., et al. 2014, ApJ, 787, 18
Ackermann, M., Ajello, M., et al. 2015, The Astrophysical
Journal, 799, 86
Ahlers, M., Bai, Y., Barger, V., & Lu, R. 2016, Phys. Rev., D93,
013009
Ahlers, M. & Halzen, F. 2014, Phys. Rev., D90, 043005
Apel, W. D. et al. 2013, Astropart. Phys., 47, 54
Arons, J. 2003, Astrophys.J., 589, 871
Bechtol, K., Ahlers, M., Di Mauro, M., Ajello, M., &
Vandenbroucke, J. 2015
Beck, R. 2012, Space Science Reviews, 166, 215
Berezinsky, V. S., Blasi, P., & Ptuskin, V. S. 1997, The
Astrophysical Journal, 487, 529
Blasi, P., Epstein, R. I., & Olinto, A. V. 2000, ApJ Letters, 533,
L123
Brunetti, G., Blasi, P., Cassano, R., & Gabici, S. 2004, MNRAS,
350, 1174
Brunetti, G. & Jones, T. W. 2014, Int. J. Mod. Phys., D23,
1430007
Brunetti, G. & Jones, T. W. 2015, in Astrophysics and Space
Science Library, Vol. 407, Astrophysics and Space Science
Library, ed. A. Lazarian, E. M. de Gouveia Dal Pino, &
C. Melioli, 557
Cavaliere, A. & Fusco-Femiano, R. 1976, A&A, 49, 137
Colafrancesco, S. & Blasi, P. 1998, Astroparticle Physics, 9, 227
Dar, A. & Shaviv, N. J. 1995, Physical Review Letters, 75, 3052
de Marco, D., Hansen, P., Stanev, T., & Blasi, P. 2006,
Phys. Rev. D, 73, 043004
Dermer, C. D., Murase, K., & Inoue, Y. 2014, arXiv e-print
1406.2633
Dermer, C. D., Murase, K., & Takami, H. 2012, ApJ, 755, 147

Fang, K., Kotera, K., Murase, K., & Olinto, A. V. 2013, arXiv
e-prints 1311.2044
Fang, K., Kotera, K., & Olinto, A. V. 2012, Astrophys.J., 750, 118
Fang, K., Kotera, K., & Olinto, A. V. 2012, The Astrophysical
Journal, 750, 118
Fang, K. & Linden, T. 2015, Phys. Rev., D91, 083501
Fang, K. & Miller, M. C. 2016
Farrar, G. R. & Gruzinov, A. 2009, Astrophys. J., 693, 329
Fujita, Y., Takizawa, M., & Sarazin, C. L. 2003, ApJ, 584, 190
Fukushima, M. 2015, EPJ Web Conf., 99, 04004
Gonzalez, A. H., Sivanandam, S., Zabludoff, A. I., & Zaritsky, D.
2013, Astrophys.J., 778, 14
Guo, F., Li, H., Daughton, W., & Liu, Y.-H. 2014, Physical
Review Letters, 113, 155005
Hong, S. E., Ryu, D., Kang, H., & Cen, R. 2014, ApJ, 785, 133
Hu, W. & Kravtsov, A. V. 2003, ApJ, 584, 702
Inoue, S., Sigl, G., Miniati, F., & Armengaud, E. 2007, ArXiv
e-prints
Jokipii, J. R. 1987, ApJ, 313, 842
Kampert, K.-H., Kulbartz, J., Maccione, L., Nierstenhoefer, N.,
Schiffer, P., Sigl, G., & van Vliet, A. R. 2013, Astroparticle
Physics, 42, 41
Kang, H., Rachen, J. P., & Biermann, P. L. 1997, MNRAS, 286,
257
Kotera, K., Allard, D., Murase, K., Aoi, J., Dubois, Y., Pierog,
T., & Nagataki, S. 2009, ApJ, 707, 370
Kotera, K. & Lemoine, M. 2008, PRD, 77, 023005
McBride, J., Fakhouri, O., & Ma, C.-P. 2009,
Mon.Not.Roy.Astron.Soc., 398, 1858
M
esz
aros, P. 2006, Reports on Progress in Physics, 69, 2259
Miniati, F. 2015, ApJ, 800, 60
Miniati, F., Ryu, D., Kang, H., Jones, T. W., Cen, R., &
Ostriker, J. P. 2000, Astrophys.J., 542, 608
Murase, K. 2015, AIP Conf. Proc., 1666, 040006
Murase, K., Ahlers, M., & Lacki, B. C. 2013, Phys. Rev., D88,
121301
Murase, K., Dermer, C. D., Takami, H., & Migliori, G. 2012, ApJ,
749, 63
Murase, K., Guetta, D., & Ahlers, M. 2016, Phys. Rev. Lett., 116,
071101

9
Murase, K., Inoue, S., & Nagataki, S. 2008, Astrophys.J., 689,
L105
Murase, K., Inoue, S., & Nagataki, S. 2008, ApJl, 689, L105
Murase, K., Inoue, Y., & Dermer, C. D. 2014, Phys.Rev., D90,
023007
Murase, K., M
esz
aros, P., & Zhang, B. 2009, Phys. Rev. D, 79,
103001
Planck Collaboration, Ade, P. A. R., Aghanim, N., Arnaud, M.,
Ashdown, M., Aumont, J., Baccigalupi, C., Banday, A. J.,
Barreiro, R. B., Bartlett, J. G., & et al. 2015, ArXiv e-prints
Rordorf, C., Grasso, D., & Dolag, K. 2004, Astroparticle Physics,
22, 167
Ryu, D., Kang, H., Hallman, E., & Jones, T. 2003, Astrophys.J.,
593, 599
Sheth, R. K. & Tormen, G. 1999, MNRAS, 308, 119
Stecker, F. W., Done, C., Salamon, M. H., & Sommers, P. 1991,
Phys. Rev. Lett., 66, 2697
Stecker, F. W., Malkan, M. A., & Scully, S. T. 2006, ApJ, 648,
774

Tinker, J., Kravtsov, A. V., Klypin, A., Abazajian, K., Warren,

M., Yepes, G., Gottl
ober, S., & Holz, D. E. 2008, ApJ, 688, 709
Vazza, F., Gheller, C., & Brggen, M. 2014, Mon. Not. Roy.
Astron. Soc., 439, 2662
V
olk, H. J., Aharonian, F. A., & Breitschwerdt, D. 1996, Space
Science Reviews, 75, 279
V
olk, H. J. & Atoyan, A. M. 1999, Astroparticle Physics, 11, 73
Werner, K., Liu, F.-M., & Pierog, T. 2006, Phys. Rev. C, 74,
044902
Winter, W. 2013, Phys.Rev., D88, 083007
Wolfe, B., Melia, F., Crocker, R. M., & Volkas, R. R. 2008, ApJ,
687, 193
Zandanel, F., Pfrommer, C., & Prada, F. 2014, MNRAS, 438, 124
Zandanel, F., Tamborra, I., Gabici, S., & Ando, S. 2015, A&A,
578, A32