Turbulence and Particle Acceleration in Giant Radio Haloes: The Origin of Seed Electrons

MNRAS 000, 1??
(2015)
Preprint 24 November 2016
Compiled using MNRAS LATEX style file v3.0
Turbulence and Particle Acceleration in Giant Radio

Haloes: the Origin of Seed Electrons
arXiv:1611.07533v1 [astro-ph.HE] 22 Nov 2016
Anders
Pinzke1,2 , S. Peng Oh3 , and Christoph Pfrommer4
1
The Oskar Klein Centre for Cosmoparticle Physics, Stockholm University, AlbaNova University Center, SE - 106 91 Stockholm, Sweden
Cosmology Center, University of Copenhagen, Juliane Maries Vej 30, DK-2100 Copenhagen, Denmark
3 University of California - Santa Barbara, Department of Physics, CA 93106-9530, USA
4 Heidelberg Institute for Theoretical Studies (HITS), Schloss-Wolfsbrunnenweg 35, 69118 Heidelberg, Germany
2 Dark
24 November 2016
ABSTRACT
About 1/3 of X-ray-luminous clusters show smooth, Mpc-scale radio emission, known
as giant radio haloes. One promising model for radio haloes is Fermi-II acceleration
of seed relativistic electrons by compressible turbulence. The origin of these seed electrons has never been fully explored. Here, we integrate the Fokker-Planck equation of
the cosmic ray (CR) electron and proton distributions when post-processing cosmological simulations of cluster formation, and confront them with radio surface brightness
and spectral data of Coma. For standard assumptions, structure formation shocks
lead to a seed electron population which produces too centrally concentrated radio
emission. Matching observations requires modifying properties of the CR population
(rapid streaming; enhanced CR electron acceleration at shocks) or turbulence (increasing turbulent-to-thermal energy density with radius), but at the expense of fine-tuning.
In a parameter study, we find that radio properties are exponentially sensitive to the
amplitude of turbulence, which is inconsistent with small scatter in scaling relations.
This sensitivity is removed if we relate the acceleration time to the turbulent dissipation time. In this case, turbulence above a threshold value provides a fixed amount
of amplification; observations could thus potentially constrain the unknown CR seed
population. To obtain sufficient acceleration, the turbulent magneto-hydrodynamics
cascade has to terminate by transit time damping on CRs, i.e., thermal particles must
be scattered by plasma instabilities. Understanding the small scatter in radio halo
scaling relations may provide a rich source of insight on plasma processes in clusters.
Key words: acceleration of particles, cosmic rays, turbulence, gamma-rays: galaxies:
clusters, radiation mechanisms: non-thermal, galaxies: clusters: general
INTRODUCTION
About one third of X-ray-luminous clusters show smooth,

unpolarised radio emission on Mpc scales, known as giant
radio haloes (RHs) (Brunetti & Jones 2014). They appear
only in disturbed, merging clusters and the RH luminosity correlates with the X-ray luminosity (Govoni et al. 2001;
Feretti et al. 2012) and the Compton y-parameter (Basu
2012; Planck Collaboration et al. 2013). The RHs show that
CR electrons and magnetic fields permeate a large volume fraction of the intra-cluster medium (ICM). The dominant CR source, given the smoothness and enormous extent of RHs, is thought to be structure formation shocks
(Miniati et al. 2001a; Pfrommer 2008). At the same time,
apinzke@fysik.su.se (AP); peng@physics.ucsb.edu

christoph.pfrommer@h-its.org (CP)
c 2015 The Authors

(SPO);
plasma processes, the origin of magnetic fields and particle acceleration in a turbulent, high- plasma (in which the
thermal pressure predominates the magnetic pressure) like
the ICM are not well understood. Radio haloes thus provide
an incisive probe of non-thermal processes in the ICM.
There have been two competing models proposed to explain RHs. The radio emitting electrons in the hadronic
model are produced in inelastic (hadronic) CR proton
interactions with protons of the ambient thermal ICM,
which generates pions that eventually decay into electrons
and positrons, depending of the charge of the initial pion
(Dennison 1980; Blasi & Colafrancesco 1999; Miniati et al.
2001b; Pfrommer & Enlin 2004; Pfrommer et al. 2008;
Enlin et al. 2011). CR protons and heavier nuclei may have
been accelerated and injected into the ICM by structure
formation shocks, active galactic nuclei and galactic winds.
However, the strong bimodality that separates X-ray lumi-
A. Pinzke, S. Peng Oh and C. Pfrommer
nous clusters into radio-active and radio-quite clusters (requiring a fast switch on/off mechanism of the RH emission)
and the very extended RH emission at low frequencies in
Coma (352 MHz) represent a major challenge to this model
class (Brunetti et al. 2012; Zandanel et al. 2014).
The alternative model for RHs is re-energisation
of seed suprathermal electrons by Fermi II acceleration when ICM turbulence becomes transsonic during
mergers (Schlickeiser et al. 1987; Giovannini et al. 1993;
Brunetti et al. 2001, 2004; Brunetti & Lazarian 2007, 2011;
Miniati 2015). Due to the short radiative cooling time of
high-energy relativistic electrons, the cluster synchrotron
emission quickly fades away after a merger, which naturally explains the observed bimodality of RHs (see e.g.
Donnert et al. 2013; Donnert & Brunetti 2014).
However, there is a salient piece missing in the turbulent
reacceleration model. It relies heavily on the assumption of
an abundant, volume-filling population of seed suprathermal
electrons; direct Fermi II acceleration from the thermal pool
is precluded by strong Coulomb losses (Petrosian & East
2008; Chernyshov et al. 2012). These seeds are presumed
to be either fossil CR electrons (CRes) accelerated by diffusive shock acceleration (DSA) during structure formation (Sarazin 1999), or secondaries injected by hadronic
interaction of CR protons (CRps) with thermal protons
(Brunetti & Lazarian 2011).
While analytic estimates have been made, there has
been no ab initio demonstration that structure formation
can lead to the required abundance of seed electrons with
the correct spatial and spectral characteristics. This is a nontrivial requirement: Coulomb cooling in dense cluster cores
is severe, and DSA fossil electrons may not survive. On the
other hand, for secondaries to constitute the seed population, the CRp population required in the best-studied case
of the Coma cluster must have a very broad and flat (or
even slightly inverted) spatial profile (Brunetti et al. 2012),
in contrast with the thermal plasma whose energy density
declines steeply with radius. In Figure 1 we show that such a
distribution is not predicted by cosmological simulations (see
also Pinzke & Pfrommer 2010; Vazza et al. 2014). If CRps
are predominantly advected with the cluster plasma, their
distribution will be peaked towards the cluster centre and
show a similar characteristics as the thermal plasma. As a
consequence, the distribution of secondary electrons and the
resulting radio synchrotron emission is also peaked since the
hadronic reaction is a two-body scattering process. Hence,
the simulated emission falls short of the observed extended
and flat radio profile of the Coma cluster.
Indeed, arriving at a seed population with the required
characteristics is highly constraining, and has the potential
to teach us much about the origin of CRps/CRes in clusters. In this work, we use our hydrodynamical zoom simulations of galaxy clusters in a cosmological setting to follow the distribution functions of seed populations for CRps
and CRes, and integrate the Fokker-Planck equation of CR
transport along Lagrangian particle trajectories. We model
diffusive shock acceleration at structure formation shocks,
and account for various loss processes of CRs. Utilising new
insights from our recent work on DSA generated fossil electrons (Pinzke et al. 2013), we generate the first quantitative
calculation of primary and secondary seed electrons.
To compare this to observations, we model second-order
10-12
Coma
CR protons
Brunetti et al. (2012); ECR=0.003 Eth
Simulation; p<0.1, ECR=0.003 Eth
CR(R) [erg/cm3]
10-13
10-14
10-15
0.1
1.0
R / R200
Figure 1. Spatial distribution of CRp energy density in the Coma

cluster. The red dash-dotted line shows the required distribution of seed CRps that generate secondary electrons via protonproton (p-p) collisions required to reproduce Coma radio brightness observations after Fermi-II reacceleration (Brunetti et al.
2012). The blue solid line shows the distribution of fossil CRps
found in cosmological simulations, which disagrees with the required profile. To better compare the two models in this figure,
we normalise the required distribution of CRps by fixing the total CRp energy ECR to 0.3 per cent of the total thermal energy,
consistent with observations (Ackermann et al. 2014; Arlen et al.
2012).
Fermi acceleration by CR interactions with magnetised turbulence. However, we assume a simplified and stationary
model for magnetic fields and turbulence. We do not account
for the time-varying energy density in compressible waves,
which are thought to be necessary for the acceleration process (Brunetti & Lazarian 2007, 2011), as the cluster merger
proceeds. So our approach is orthogonal (and complementary) to e.g., simulations of Miniati 2015 that focus on the
time-dependent compressible turbulence while adopting a
simplified treatment of CR. Our approach of parametrizing
turbulence enables us to vary parameters associated with
the spatial profile and the overall amplitude of compressible
waves (that can in principle vary depending on the details
of a particular cluster merger).
In this paper, we explore how the radio surface brightness profile and spectrum of the best known radio halo,
Coma, can be used to constrain the underlying properties
of the seed CRs and turbulence. We aim to constrain the
normalisation and spatial profile of these two input ingredients in turbulent reacceleration models. The outline of
this paper is as follows. In Section 2, we outline the basic physics of turbulent reacceleration of CRs which we use.
In Section 3, we use cosmological simulations to generate a
seed CR population, and combine it with our parametrized
model of turbulence to produce radio surface brightness profiles and spectra of Coma. We find that it is possible to fit the
observations using physically motivated modifications of the
seed population (rapid streaming; enhanced CR electron acceleration at shocks) or turbulence (increasing the turbulentto-thermal energy density, turb /th with radius), but only at
the expense of fine tuning. In Section 4, we explore the reason for this fine-tuning, and seek ways to overcome it. We
MNRAS 000, 1?? (2015)
Origin of Seed Electrons

perform a parameter study on spherically symmetric, static
models where we vary properties of the seed population and
turbulence. We find exponential sensitivity to the amount of
turbulence, which can be eliminated if the turbulent acceleration and dissipation time are linked. In this case, turbulence above a threshold value required to overcome cooling
provides a fixed amount of amplification; observations could
then potentially constrain the unknown CR seed population.
We summarise and conclude in Section 5.
COSMIC RAY TRANSPORT
The transport of relativistic electrons and protons across

cosmic time into galaxy clusters is a complex problem that
depends on the velocity field of the gas (and its thermodynamic properties such as density, temperature, and pressure) as well as non-thermal processes (turbulence, magnetic fields, fossil CRs). We use high resolution galaxy
cluster simulations to derive the thermal and fossil CR
properties (shock accelerated primary CRes and CRps, as
well as secondary CRes produced in p-p collisions, see
Pfrommer et al. 2007; Pfrommer 2008; Pinzke & Pfrommer
2010; Pinzke et al. 2013).
2.1
Basic equations
As previously noted, secondaries produced by shock accelerated CRp have the wrong spatial profile to explain
RH observations. Because they arise from a two body process, they are too centrally concentrated. They also produce gamma-ray emission in excess of Fermi-LAT upper limits (Arlen et al. 2012; Brunetti et al. 2012; Ackermann et al.
2014; Ahnen et al. 2016).
Given a seed population of CRs, we adopt essentially
the same set of plasma physics assumptions as the reacceleration model for RHs (Brunetti & Lazarian 2007, 2011).
We solve the isotropic, gyro-phase averaged Fokker-Planck
equation (via a Crank-Nicholson scheme) for the time evolution of the CRe distribution in the Lagrangian frame
(Brunetti & Lazarian 2007, 2011):
d fe (p, t)
dt
(
"

dp
dp
p
+ ( 3) +
fe (p, t)
p
dt Coul 3
dt rad
#)
1 2
( 3) fe (p, t)
p D pp
p2 p
i
h
i
2 h
D pp fe (p, t) + Qe p, t; fp (p, t) .
(1)
2
p
Here fe is the one-dimensional distribution in position x

(suppressed for clarity), momentum p = P/(me c) and time
t (which is normalised
such that the number density is given
R
by nCRe (t) = d p fe (p, t)), d/dt = /t + 3 is the Lagrangian
derivative, 3 is the gas velocity, |d p/dt| represents Coulomb
(Coul, Gould 1972) and radiative (rad, Rybicki & Lightman
MNRAS 000, 1?? (2015)
1979) losses, respectively,

d p
dt Coul
3 T ne c me c2 e 1
ln
2 2e ~ plasma
!
!2
2 1
1
1
ln(2) e +
+ +
,
2
2
4
!2
4 T p2
B
CMB .
1 +
3 m e c e
BCMB
(2)

d p
=
(3)
dt rad
p
of CRs, =
Here e = p/ 1 + p2 is the dimensionless velocity p
p
1 + p2 is the Lorentz factor of CRs, plasma = 4e2 ne /me
is the plasma frequency, ne is the number density of free
electrons, and T = 8e4 /3(me c2 )2 is the Thomson cross section. The rms magnetic field strength is denoted by B and
the equivalent field strength of the cosmic-microwave background is given by BCMB = 3.24(1+z)2 G, where z denotes the
redshift. In the peripheral cluster regions, where B BCMB ,
the CRes loose virtually all their energy by means of inverse
Compton emission. D pp is the momentum space diffusion
coefficient (see Section 2.2), and Qe denotes the injection
rate of primary and secondary electrons in the ICM (see
section 3.1). The first term containing the expression 3
represents Fermi-I acceleration and the second term of this
form describes adiabatic gains and losses.
During post-processing of our Coma-like cluster simulation, we solve the Fokker-Planck equation over a redshift interval from z = 5 to 0. The simulated cluster undergoes a major merger over the last 1-2 Gyrs that is
thought to inject large turbulent eddies. As is commonly
assumed (Brunetti & Lazarian 2007, 2011; Yan & Lazarian
2004; Beresnyak et al. 2013) we assume that about one Gyr
after core passage the fields have decayed down to the smallest scales kcut , and the radio halo turns on shortly after. We
choose this simulation snapshot to analyse. Note that in recent simulations by Miniati (2015), the turbulent reacceleration is strongest around core passage. However, we are not
very sensitive to the adopted decay time, since the thermal
and CR quantities are very similar a few 100 Myrs before
and after z = 0, where we have chosen to evaluate the simulations. In all our calculations we assume that turbulent reacceleration efficiently accelerates particles for cl 650 Myrs
(which is roughly the cascade time on which turbulence is
damped) and that during this turbulent phase CR streaming and spatial diffusion can be neglected. In Section 4.3, we
explore sensitivity to the last assumption.
Thus far, we have ignored CR transport. However,
if CRps stream in the ICM, then their spatial profile
could potentially flatten sufficiently (Enlin et al. 2011;
Wiener et al. 2013). This scenario is very attractive: it generates seed electrons with the right spatial footprint, and
by removing CRps from the core, obeys gamma-ray constraints. Turbulence plays two opposing roles: Alfvenic turbulence damps waves generated by the CR streaming instability (Yan & Lazarian 2002; Farmer & Goldreich 2004),
thus reducing self-confinement; but compressible fast modes
scatter CRs directly. Turbulent damping is still efficient for
highly subsonic conditions (Wiener et al. 2013), while we
assume compressible fast modes to only provide effective
spatial confinement during the periods of transsonic, highly
super-Alfvenic (MA 5) turbulence associated with mergers. Thus, CRs can stream out when the cluster is kine-
matically quiescent. Furthermore, even Alfvenic streaming

timescales are relatively short ( 0.1 0.5 Gyr; Wiener et al.
2013) compared to the timescale on which the CRp population is built up. Based on these findings, we adopt a toy
model for our M-streaming scenario in which CR streaming
instantaneously produces flat CRp profiles. We assume that
CRs cannot stream significantly past perpendicular B-fields
at the accretion shock, so that the total number of CRs is
conserved within the virial radius during the streaming process.
The time evolution of the spectral energy distribution
of CRps, fp , is similarly given by:
(
"
d fp (p, t)
p
dp
+ (3 + 3st )
=
fp (p, t)
dt
p
dt Coul 3
#)
1 2
p D pp
fp (p, t) (3 + 3st ) 3st fp (p, t)
2
p p
i fp (p, t)
2 h
+
D pp fp (p, t)
+ Qp (p, t) ,
(4)
2
p
had (p)
where 3st = 3A fp /| fp | is the streaming velocity in the
isotropic transport approximation, 3A is the Alfven speed,
and Qp (p, t) denotes the injection rate of shock accelerated
CRps as a function of momentum p = P/(mp c) and time t (see
section 3.1). The timescale of hadronic losses that produce
pions via CRp collisions with thermal protons of the ICM is
given by
h
i1
had = c nth +/,0 (p) ,
(5)
where we use the cross-section, +/,0 (p), given by the fitting

formula in Dermer (1986).
2.2
Turbulent reacceleration
In turbulent reacceleration, particles gain energy via Fermi

II acceleration. Wave-particle energy exchange takes place
via transit time damping (TTD, Brunetti & Lazarian 2007,
2011). The TTD resonance requires the wave frequency to
obey the resonance condition, = kk 3k , where kk and 3k are
the parallel (projected along the magnetic field) wavenumber
of the compressible mode and particle velocity, respectively.
This implies that the particle transit time across the confining wave region matches the wave period, k /3k = wave (hence
the moniker, transit time damping). Note that the CRs gyroradius does not enter the resonance condition. Hence the
CRs that are in resonance with compressible waves experience Fermi-II acceleration irrespective of the length scale of
the perturbation.
However, the resonance changes the component of the
particle momentum parallel to the mean magnetic field,
which over time leads to increasing anisotropy in the particle distribution that decreases the efficiency of reacceleration with time. As in Brunetti & Lazarian (2011), we assume that there exists a mechanismsuch as the firehose
instabilitythat isotropises the CR distribution function at
the gyroscale and on the reacceleration time scale, which
ensures sustained efficient reacceleration with time.
All of the physics of turbulent reacceleration is
effectively encapsulated in the diffusion coefficient
(Brunetti & Lazarian 2007), which can be rewritten as
(Miniati 2015):
p2 I (cs /c)
hkiW h(3c )2 i
(6)
8c
where I averages interaction rates over the CR pitch angle
; I 5 for ICM conditions, and
!2s
Z kcut
kcut
s1
1
k
dk
k
W(k)
(7)
hkiW =
L
h(3c )2 i kL
2s
kL
D pp (p) =
is an energy-averaged wavenumber, kL , kcut are the wavenumbers associated with the outer scale L and the cutoff scale
respectively, and we have assumed a total energy spectrum
(composed of both kinetic and potential energy, where the
two are assumed to be in equipartition, Sarkar et al. 1991):
!s
(s 1)h(3c )2 i k
W(k) =
(8)
kL
kL
which defines the normalisation h(3c )2 i (the subscript c
emphasizes that we specialise to compressive modes). Intuitively, we can understand the form of the diffusion coefficient from the fact that for second-order Fermi acceleration,
p p/ p hkciW (3c /c)2 , where 1 hkciW is the energy
averaged wave-particle interaction rate, and p p(3c /c)2
is the typical momentum change during wave-particle scattering. Thus:
3 2
c
.
(9)
D pp p p p2 hkciW
c
Equations (6) and (8) make the important aspects of

turbulence clear: the energy in compressive modes h(3c )2 i,
the inner and outer scale (kcut and kL ), and the slope of the
energy spectrum s. All of these can vary spatially and temporally.
We adopt a Kraichnan spectrum (s = 3/2) for the fast
modes, as seen in simulations (Cho & Lazarian 2003). This
can also be written as:
q
W(k) = 2/7 IL h3ph i k3/2 ,
(10)
where IL is the volumetric injection rate of turbulence at

scale L (which is assumed to be constant), 3ph cs is the
phase speed of waves. Since the Kraichnan spectrum is critical in what follows, it is worthwhile taking a moment to
recount the origin of the spectral slope s and the cascade
rate. In the magnetically dominated regime, two counterpropagating wave packets of scale l interact on a wave crossing time ph = l/3ph , rather than the eddy turnover time
edd = l/3l . Since ph edd , each interaction results in a
small velocity change 3l 3l (ph /edd ). If these changes behave like a random walk, the cascade time l it takes for an
eddy to become non-linear (3 3l ) and cascades to smaller
scales occurs with a characteristic velocity 3l 3l (l /ph )1/2
3l (ph /edd ) (l /ph )1/2 , which we can solve to obtain the cascade time:
!2
l3ph
edd
l
ph 2
(11)
ph
3l
where we have implicitly assumed isotropy. This implies a
dissipation rate 32l /l 34l /(l3ph ), or 3l (l3ph )1/4 . Thus,
since kE(k) 32l , this gives a kinetic energy spectrum:
E(k) (3ph )1/2 k3/2 .
(12)
It is important to realise that this Kraichnan scaling

MNRAS 000, 1?? (2015)

only applies at small scales, where turbulence is magnetohydrodynamical. At large scales, turbulent motions are subsonic (Ms 0.2 0.5) but super-Alfvenic (MA 5). Thus,
except in the case where motions are transsonic and weak
shocks become important, motions are fundamentally hydrodynamic and turbulence follows a Kolmogorov (s =
5/3) spectrum. However, while the turbulent energy density turb l2/3 decreases at small scales, the magnetic energy density (which is dominated by the large scale mean
field) is scale-independent. Thus, at some scale lA lL M3
A ,
where turb B , the magnetic field becomes dynamically
important, and turbulence transitions to the MHD regime.
The cutoff scale in equation (8) can be computed by setting the turbulent cascade rate for fast modes to the transit time damping rate on thermal electrons. This yields
(Brunetti & Lazarian 2007; Miniati 2015):
kcut Akouter
h(3outer )2 i2
c4s
(13)
where A 11000. If there is an unimpeded cascade from the

injection to the cutoff scale, kouter = kL and 3outer = 3c . However, in our case the hydrodynamic (Kolmogorov) cascade
transitions to the MHD cascade at the Alfven scale so that
kouter = kA and 3outer = 3A and hence, we get1
kcut AkA 2 .
(14)
This gives 2/kcut 0.1 1 kpc in the ICM. This constitutes an effective mean free path for CRs, unless plasma instabilities can mediate interactions between turbulence and
particles on smaller scales (Brunetti & Lazarian 2011), a
possibility we discuss in Section 4.3. Another possibility is
that compressible modes dissipate in weak shocks, resulting in Burgers turbulence s = 2 (Kowal & Lazarian 2010;
Porter et al. 2015; Miniati 2015). If Burgers turbulence dominates, then particle acceleration rates are too slow in the
face of cooling processes to explain radio haloes (Miniati
2015), and an alternative model for radio haloes is required.
The spatial profile of injected turbulence depends on
the details of the merger such as time during the merger,
the impact parameter, the merger mass ratio, and the degree
of cluster anisotropy (Miniati 2015). We parametrize these
uncertainties and assume that volumetric injection rate of
turbulent energy, IL thtu , and determine the normalisation by requiring
R R that the turbulent energy in compressible
modes Eturb =
W(k)dkdV = Xtu Eth , where Eth is the total
thermal energy. Given these definitions, one can show that:
D pp
IL
tu 1
Xtu2 kL th
T.
cs
(15)
What is a typical energy density in compressive fast

modes? The total turbulent energy is typically 15 70 per
cent of the thermal energy in a cluster (Vazza et al. 2011a);
it rises rapidly during a merger. The compressible component is 20 40 per cent of the total turbulent energy,
and shows more rapid temporal variations compared to the
incompressible component (Beresnyak et al. 2013; Miniati
Note that this differs from previous work, which assumes

Kraichnan turbulence from the injection scale onward and effectively adopts wave number and compressible velocity at the outer
scale L for this estimate of kcut .
MNRAS 000, 1?? (2015)
2015). Note the compressible component in cluster simulations is much larger than in stirring box simulations with
similar Mach numbers (Kowal & Lazarian 2010; Lynn et al.
2014). This is likely due to the compressive nature of turbulent driving (transsonic infall and merger), whereas idealised simulations tend to use incompressible solenoidal driving and allow compressive fluctuations to develop on their
own. Overall, we adopt a compressive energy density which
is Xtu 0.2 of the thermal energy as a baseline estimate.
The important effects are best summarised in terms of
the acceleration rate which is governed by advection in momentum space:
1
D D
4D pp
p
2
= p3
.
p D pp =
p
p
p2
(16)
In the last step we have used that D pp p2 for turbulent reacceleration (equation 9). Hence the acceleration time is independent of momentum. This should be
compared against the lifetime of turbulence, cl , and the
cooling time cool . In Table 1, we show both the thermal quantities and the timescales for CR cooling and
(re)acceleration for three different spatial regions of the
RH. The densities (Briel et al. 1992) and temperatures
(Bonamente et al. 2009; Arnaud et al. 2001) are derived
from X-ray observations. To calculate synchrotron cooling
times, we use B-fields derived from Faraday rotation measurements (Bonafede et al. 2010). To calculate the acceleration time, we need to assume an outer scale. We assume an
injection scale kL = 2/L , where L = 100 kpc, which were
assumptions adopted in previous work (Subramanian et al.
2006; Brunetti & Lazarian 2007, 2011). This length scale
corresponds to an eddy turnover time on the outer scale
1
of 2L 31
L 1.2 Gyr if 3L 500km s , as is characteristic of
a merger. Note that hydrodynamical simulations of clusters
have sometimes found larger L , in some cases comparable
to the size of the cluster (e.g. L 1Mpc in Vazza et al.
2011a; Miniati 2015). This choice is degenerate with Xtu ; in
Section 4.3 we argue that kA is a more appropriate choice,
but also find that when the decay time is appropriately
scaled, we are relatively insensitive to the choice of kL . We
present D for 3 different models, which we discuss in the
next section. The reacceleration timescale D is similar between our three models, where the difference comes from
turbulent profile parametrized with tu . This implies that
even small differences in the turbulent profile could impact
the seed CRs significantly. Finally, we adopt an duration of
acceleration cl = 650 Myr, in line with previous assumptions (Brunetti & Lazarian 2007), roughly corresponding to
the turbulent decay time.
COSMOLOGICAL SIMULATIONS
In this section, we solve the Fokker-Planck equation for CR

transport on Lagrangian particle trajectories through cosmic
history. We then use our parametrized model of turbulence
to apply turbulent re-acceleration, and compare radio halo
profiles against observations of the Coma cluster. We follow
the philosophy of adopting assumptions roughly in line with
previous successful models, which are now confronted with
more accurate calculations of the seed CR population, and
Table 1. Thermal quantities and timescales for different spatial

regions in a Coma like cluster.
thermal quantities(1)
[1027 g cm3 ]
T [108 K]
timescales(1)
D (M-primaries)(3) [Gyr]
D (M-streaming)(3) [Gyr]
D (M-turbulence)(3) [Gyr]
IC/sync (P = 104 me c)(4) [Gyr]
had (P = 100 mp c)(5) [Gyr]
Coul (P = me c)(6) [Gyr]
0.1 R(2)
RH
spatial regions
0.3 R(2)
RH
R(2)
RH
3.0
1.4
1.6
1.0
0.15
0.58
0.45
0.50
0.69
0.11
2.4
0.0092
0.44
0.47
0.56
0.15
4.5
0.017
0.39
0.34
0.27
0.22
47
0.17
find the minimal modifications required to fit observations.

We re-examine these choices in Section 4.
Since the standard vanilla model requires a CR seed
population which is inconsistent with the simulations (Figure 1), some modifications are necessary, either in the seed
CR population or in the properties of turbulence. We focus on three scenarios. (i) In model M-primaries, we assume
that CR electrons, which have been accelerated by cosmic
formation shocks and successively cooled by inverse Compton and synchrotron losses, form a fossil seed population for
reacceleration. (ii) In model M-streaming, we account for
the outward streaming of central CRps, which produces a
flat CR distribution in the ICM and equivalently a flat secondary seed population of CRes for reacceleration. (iii) In
model M-turbulence we adopt a spatially flatter turbulent
profile than what was adopted before but assume that seed
CRps and secondary CRes follow the steep profile that is
suggested by structure formation simulations.
Modelling diffusive shock acceleration
In this paper we focus on our simulated cluster, g72a, which

is a massive cluster of mass M200 = 1.6 1015 M that experienced a merger about 1 Gyr ago (Dolag et al. 2009). Since
the cluster mass, density and temperature profiles are all
similar to the well studied Coma cluster (Pfrommer et al.
2007; Pinzke & Pfrommer 2010), we will compare our calculations to radio and gamma-ray observations of Coma.
We use a simple test-particle model for the CR acceleration and injection, where each shock injects CRs that trace
a power-law in momentum,
fp (p, t) = C(t) pinj ,
inj =
pinj
CR
Notes:
(1) Median quantities from our simulated post-merging cluster
g72a derived during last 300 Myrs in time.
(2) Radius of the giant radio halo in Coma where RRH 0.6 R200 .
(3) Fermi-II reacceleration for both electrons and protons at all
energies.
(4) Inverse Compton and synchrotron cooling for electrons.
(5) Catastrophic losses for protons.
(6) Coulomb cooling for electrons (protons factor me /mp smaller).
3.1
inj that depends on the adiabatic index ad = 5/3 and the

Mach number of the shock M (see also Quilis et al. 1998;
Miniati et al. 2001a; Pfrommer et al. 2006). It is given by
the ratio of the upstream velocity (32 ) and the sound speed
(cs ). The CR number density and CR energy density are
derived from
Z
nCR =
dp fp (p)
(18)
(ad + 1)M2
(ad 1)M2 + 2
(17)
determined by the normalisation C(t) and the spectral index
dp fp (p) E(p),
(19)
pinj
p
where E(p) = ( 1 + p2 1) mp c2 is the kinetic energy of a
proton with momentum p. We adopt a fit to Monte Carlo
simulations of the thermal leakage process that relates the
momentum of injected protons (pinj ) to the thermal energy
(pth ) of the shocked plasma (Kang & Ryu 2011):
s
2 kB T 2
,
pinj = xinj pth = xinj
mp c2
!0.1
!
32
1.07 M
where xinj 1.17
1+
.
(20)
pth c
B
3
Here B = B0 /B , B0 is the amplitude of the downstream
MHD wave turbulence, and B is the magnetic field along
the shock normal. The physical range of B is quite uncertain
due to complex plasma interactions. In this paper, we adopt
B = 0.23, which as we will later see corresponds to a conservative maximum energy acceleration efficiency for protons
of 0.1. To derive the acceleration efficiency, inj , we first have
to infer the particle injection efficiency, which is the fraction
of downstream thermal gas particles which experience diffusive shock acceleration (for details see Pinzke et al. 2013),
3
xinj
4
x2
e inj .
p,lin =
1
inj
(21)
The particle injection efficiency is a strong function of xinj

that depends on both M and B . The energy density of CRs
that are injected and accelerated at the shock (neglecting
the CR back reaction on the shock) is given by
CR
CR,lin = p,lin (M) nth (T 2 )
(22)
nCR
and the CR energy injection and acceleration efficiency is:
!
CR,lin
2 ad
.
(23)
, where diss = th2 th0
lin =
diss
0
The dissipated energy density in the downstream regime,
diss , is given by the difference of the thermal energy densities in the pre- and post-shock regimes, corrected for the
adiabatic energy increase due to gas compression.
We limit the acceleration efficiency to max by steepening the spectral index of the injected population inj to sub
so that lin 6 max is always fulfilled. The slope inj impact
inj via the mean energy per particle, p /np . This procedure
conserves energy and is motivated by models of non-linear
shock acceleration where a sub-shock with a lower compression ratio (and hence steeper spectral index) forms (e.g.,
Ellison et al. 2000). Given our assumed B = 0.23, we find
that for strong shocks where inj . 2.3 the spectral slope is
steepened by a maximum of 10 per cent in low temperature regimes (kB T 0.1 keV), while the steepening is much
MNRAS 000, 1?? (2015)

smaller for high temperature regimes (kB T 10 keV) that
are more relevant for clusters. Since pinj remains fixed, so
does the CR number density nCR . Hence we can solve for
the renormalised normalisation constant Csub using nCR and
Eqn. 21:
sub 1
Csub = p,lin (sub 1) pinj
,
(24)
where the new distribution function is given by fp (p, t) =

Csub psub . We set an upper limit on the ratio of accelerated
proton-to-dissipated energy in the downstream of strong
shocks that varies from max 0.01 0.1, depending on the
adopted model (for more details, see section 3).
In our Galaxy, the CRe-to-CRp ratio at a few GeV
is Kep 102 . Hence, we adopt this as a fiducial value
for the CRe-to-CRp acceleration efficiency in our models M-streaming and M-turbulence (see Pinzke et al. 2013,
for more discussion). However, as recent PIC simulations
have shown, this is likely very different at weak shocks,
with electrons efficiently accelerated at perpendicular shocks
(Guo et al. 2014a,b), and ions (and electrons) efficiently accelerated at parallel shocks (Caprioli & Spitkovsky 2014;
Park et al. 2015). Thus, depending on magnetic geometry,
Kep could be either larger or smaller. Some observations of
radio relics suggest high values of Kep , due to the absence
of gamma-ray emission, which probes the CRp population
(Vazza & Br
uggen 2014). This suggests primary CRes as a
viable alternative scenario to secondary CRes as seeds for the
giant RHs. In our M-primaries scenario, the injected distribution of CRes is derived in the same way as for the CRps.
Once they have been accelerated to relativistic energies, injected electrons and protons are indistinguishable. We therefore assume that CRp and CRe have the same distribution
function fe (p) = Kep fp (p), with a different normalisation (due
to differing acceleration efficiencies) Kep = 0.1 (which is viable for primarily perpendicular shocks, Guo et al. 2014a).
Given the unknown magnetic geometry at cluster shocks,
we investigate the consequences of this additional degree of
freedom.
3.2
Radio emission profile
In Figure 2, we show radial profiles for the radio emission

in all three scenarios in which the seeds undergo Fermi-II
reacceleration in turbulent fields that are shaped such as to
reproduce the Coma RH profile at 352 MHz. Adopting our
parametrization for the volumetric injection rate of turbulent energy, IL thtu , we find that to fit the observations,
we require tu = 0.67 for M-turbulence, tu = 0.82 for Mstreaming, and tu = 0.88 for M-primaries. As a result, the ratio of turbulent-to-thermal energy densities slightly increase
with radius as shown in Figure 3. The fine-tuning of these
exponents is somewhat problematic, as we discuss in Section 4. After turbulent reacceleration, the volume-weighted,
relative CRp energy density and relative CRp number density inside the RH for M-turbulence (M-streaming), are found
to be 2 (3) per cent and 2 108 (5 108 ), respectively.
Figure 2 demonstrates that the modelled radio profiles
without turbulent reacceleration are too steep. In the bottom right panel of Figure 2 (labelled with Brunetti et al.
2012), we show that our simulated profiles of reaccelerated
CRs, which only take advective CR transport into account,
MNRAS 000, 1?? (2015)
i.e. they neglect CR streaming or a flatter turbulent profile, produce radio profiles that are too steep. Indeed, even
using the assumptions of previous work, where complete
freedom in the seed population was allowed, it is not possible to reproduce observations in both frequencies in any
model2 with or without turbulent reacceleration. Decreasing the acceleration efficiency with radius does not change
this conclusion much because
of the weak radial dependence
of D pp (R) th (R)tu 1 T (R). This signals that the problem

is generic and requires either additional modifications to the
plasma physics of acceleration or a better understanding of
potential observational systematics. In addition there are
differences in the simulated density and temperature profiles
in comparison to the observed profile in Coma that impact
the CR abundance as well as cooling and reacceleration.
In Figure 4 we show how the turbulent reacceleration
timescales in our three models scale with radius. As expected, the M-primaries model with tu = 0.88 has the flattest profile with D 0.4 Gyr, where the small dip at large
radius driven by the decrease in thermal energy density.
The M-turbulence model has the flattest turbulent profile
parametrized by the smaller tu which explains the steepest D profile. Note that a fixed reacceleration timescale is
required to explain the observations at each radius and for
each model (see also Table 1). Since cl Xtu2 kL , these two parameters are degenerate and can be traded off one another.
In principle, reacceleration via TTD leads to spectral
steepening with particle energy due to the inefficiency of the
acceleration process to counter the stronger cooling losses
with increasing energy. Since synchrotron emission peaks at
frequency syn 1 B/G(/104 )2 GHz, this translates into a
spectral steepening of the radio spectrum (see the left panel
of Figure 5 where the continuous injection of secondary CRes
is absent). A given radio window samples higher energy electrons for a decreasing field strength in the cluster outskirts.
Hence, the spectral steepening with energy should translate into a radial spectral steepening (Brunetti et al. 2012).
However, because of the weak dependence of the electron
Lorentz factor on emission frequency ( syn ), this effect

is only visible in our simulations for syn & 5 GHz. Most
importantly, our simulated fluid elements at a given radius
sample a broad distribution of shock history, density and
temperature, which implies very similar synchrotron brightness profiles at syn = 352 MHz and 1.4 GHz. The discrepancy of the observed and simulated 1.4 GHz profiles could
instead be due to systematic flux calibration error in single
dish observations. These could arise, for instance, due to errors in point source subtraction. Interestingly, we can match
the 1.4 GHz data if we reduce the zero point by adding
0.1 of the central flux to every data point; this flattens the
outer profile3 . Alternatively, this may point to weaknesses
in the theoretical modelling of the particle acceleration process and may require a stronger cutoff in the particle energy
spectrum.
Note that in previous work on the Coma cluster, turb

th was adopted which approximately corresponds to tu = 1
(Brunetti et al. 2012) and together with the different distributions for seed CRes constitute the main differences compared to
our work.
3 Lawrence Rudnick, private communication.
M-streaming:
M-primaries:
10-2
10-1 secondary CRe
DSA+Fermi II reacc.
DSA
352 MHz
1.4 GHz
1.4 GHz zero level
S [Jy arcmin-2]
S [Jy arcmin-2]
10-1 primary CRe
10-3
10-4
10-5
10-6
10-2
10-3
10-4
10-5
10-6
0.1
1.0
0.1
R T / R 200
Brunetti et al. (2012):
10-2
DSA+Fermi II reacc.
DSA
352 MHz
1.4 GHz
1.4 GHz zero level
10-1 secondary CRe Simulations; DSA+FII reacc.

Brunetti et al. (2012)
S [Jy arcmin-2]
10-1 secondary CRe
10-3
10-4
10-5
10-6
10-2
10-3
10-4
10-5
-6
10
0.1
1.0
R T / R 200
M-turbulence:
S [Jy arcmin-2]
DSA+Fermi II reacc.
DSA
352 MHz
1.4 GHz
1.4 GHz zero level
1.0
R T / R 200
352 MHz
1.4 GHz
1.4 GHz zero level
0.1
1.0
R T / R 200
Figure 2. Radio surface brightness profiles of Fermi-II reaccelerated CR electrons of a simulated post-merging cluster similar to Coma.
We compare profiles at 352 MHz (blue lines and crosses, Brown & Rudnick 2011) to those at 1.4 GHz (green lines and crosses, Deiss et al.
1997). The red crosses show the reprocessed 1.4 GHz data, where a zero level of about 0.1 of the central value is adopted. The solid
lines show predicted emission from a reaccelerated fossil population, while dotted lines show emission from a fossil population without
reacceleration. The panels show the emission of our models M-primaries (upper left panel), M-streaming (upper right panel), M-turbulence
(lower left panel), and simulated secondary electrons together with previous estimates (Brunetti et al. 2012) for the Coma cluster (lower
right panel).
3.3
Radio spectrum
In Figure 5 we show that our three models that include

Fermi-II reacceleration can individually reproduce the convexly curved total radio spectrum found in the Coma cluster.
Seed CRs in M-streaming and M-turbulence that do not experience turbulent reacceleration have a power-law spectrum
in disagreement with observations. In order to match both
the spatial and spectral profiles in Coma, we adopt an acceleration efficiency for the strongest shocks in our three models
M-primaries, M-streaming, and M-turbulence to e < 0.003,
p < 0.1, and p < 0.03, respectively. Following the Mach
number (M)-dependence of the acceleration efficiency suggested in Pinzke et al. (2013), the efficiency in weak shocks
(M 2.5 3.5) that dominates the CR distribution function,
has an acceleration efficiency for protons p 0.0001 0.01,
and for electrons e 0.001.
Interestingly, we find that the radio luminosity from
clusters in the OFF-state (DSA only) and ON-state (DSA

and reacceleration) differ by about a factor 10-20 in all our
three models. This means that the secondary CRes are dominated by the reaccelerated fossil CRes and not from the
CRes produced by reaccelerated CRps. However, for high
frequencies (syn & GHz) where synchrotron cooling is more
efficient than reacceleration, the emission is dominated by
the CRes produced in the continuous injection of electrons
from reaccelerated CRps. It is also worth mentioning that
the radio emission from secondary CRes are smoothly distributed around the cluster because of the continuous injection, hence it is not dominated by outliers.
However, for M-primaries, the primary CRes that generate most of the radio emission from the cluster in the OFFstate are dominated by only a small fraction of the CRes.
These electrons are injected very recently and have not had
time to cool yet. Hence we expect there to be a large variance in the OFF-state of different simulated clusters. As
MNRAS 000, 1?? (2015)
turb(R) / th(R), Eturb(< R) / Eth(< R)

0.6
reader to take this limit too stringent because of the uncertainty in kL and cl that impact Xtu for a fixed D . This
parameter space needs to be explored further in future work
in order to put more stringent limits on the level of turbulence in clusters using radio and gamma-ray observations in
combination with turbulent reaccelerated CRs.
Compressional turbulence
0.5
Eturb(< R) / Eth(< R)
turb(R) / th(R)
0.4
M-primary (tu=0.88)
M-streaming (tu=0.82)
M-turbulence (tu=0.67)
0.3
3.4
0.2
0.1
RRH
0.0
0.2
0.4
0.6
R / R200
0.8
1.0
Figure 3. The ratio of turbulent-to-thermal energy densities (solid lines) and cumulative energies (dotted lines) in our
three models. The energy densities are parametrized as turb
(tu +1)/2 1/4
th
T
and normalised such that the total turbulent energy in compressible modes Eturb for each scenario makes up about
20 per cent of the total thermal energy Eth inside the radio halo
(RRH 0.6R200 ). The turbulent profiles explore the uncertainty
in the cluster turbulence and are motivated by the cosmological simulation in (Lau et al. 2009; Shaw et al. 2010; Vazza et al.
2011b).
0.8
M-primary (tu=0.88)
M-streaming (tu=0.82)
M-turbulence (tu=0.67)
0.7
0.6
D [Gyr]
0.5
0.4
Gamma rays
The gamma-ray emission from CRps that produce decaying

neutral pions could be substantial if the CRps are reaccelerated efficiently enough, hence it is interesting to estimate
this emission for our models and compare to upper limits.
We follow the formalism outlined in Blasi & Colafrancesco
(1999) (and references therein) and calculate the gamma-ray
emission numerically for our three models. We predict the
gamma-ray emission from M-turbulence (M-streaming) with
F (> 500 MeV) = 4 1010 (5 1010 ) ph s1 cm2 . The fluxes
from these models are slightly larger than in Brunetti et al.
(2012), where the differences comes from our steeper CRp
profiles in addition to the simulation based formalism we
rely on that accounts for both Coulomb and hadronic losses
during the build up of the CR distribution in contrast to
the scaling relations adopted in their paper. Interestingly
the gamma-ray flux from both our scenarios are just below recent Fermi-LAT limits derived from a gamma-ray
profile similar to M-turbulence4 where F (> 500 MeV) <
5.3 1010 ph s1 cm2 . The spectral index of the CRp distribution is relatively steep (p 2.6) for the CRp energies
E & 10 GeV that are relevant for the injection of radioemitting secondary CRes. The steep spectrum is ultimately
a consequence of the shock history of the simulated cluster, with a weak dependence on our test particle model for
Fermi-I acceleration (Pinzke et al. 2013), where we steepen
the spectral index to avoid acceleration efficiencies above
p 0.1.
0.3
4
0.2
0.1
0.1
1.0
R / R200
Figure 4. Turbulent reacceleration timescales for our simulated

cluster g72a. We show in linear-log the reacceleration timescale
(D ) as a function of radius R for our three models: M-primaries
(green dotted line), M-streaming (blue dashed line), and Mturbulence (red solid line). Note that the timescales are derived
during the last 100 Myrs in time in our simulations.
mentioned in section 4, combining radio observations with

gamma-ray limits allows us to put a lower limit to Xtu . If
Xtu is smaller than in our adopted models (where we assume
Xtu = 0.2), then the efficiency of DSA has to be larger than
p 0.1 for the secondary CRes to reproduce the radio observations. However, since the turbulent reacceleration acts on
both the secondary CRes and the CRps, while p only affects
the CRps, M-streaming and M-turbulence would produce too
much gamma-rays. Hence we conclude that Xtu & 0.2 if all
other parameters are kept fixed. Although, we caution the
MNRAS 000, 1?? (2015)
PARAMETER SPACE EXPLORATION AND

OVERCOMING FINE TUNING
In this paper we rely on several critical parameters describing relatively unknown non-thermal physics in the ICM.
Here we develop a simplified framework for our reacceleration model of secondary electrons. We will explore how radio
emission depends on the parameters describing the spatial
profile of CR protons and turbulence. Our fiducial model is
meant to be compared against the Coma cluster.
4.1
Methodology
As we have seen before, the most uncertain aspects of radio

halo emission models are the profile of compressible turbulence (which determines the amount of Fermi II acceleration)
and the distribution of pre-existing CRs. Hence we vary parameters describing the amount of energy contained in turbulence, Xtu (defined by Eturb = Xtu Eth ), the spatial profile
4
Fabio
Zandanel,
private
communication;
see
also
Zandanel & Ando (2014); Ackermann et al. (2014); this will
be probed in the next few years by Fermi-LAT.

1034 Primary CRes
1034 Secondary CRes
1033
1033
L () [erg/s/Hz]
L () [erg/s/Hz]
10
1032
1031
1030
M-primaries, e<0.003
1029
10
[Hz]
10
1031
1030
1029
ON-state: DSA+Fermi II reacc.

OFF-state: DSA
Pizzo et al. 2010
1028
107
1032
10
10
1028
107
M-streaming, p<0.1
M-turbulence , p<0.03
ON-state: DSA+Fermi II reacc.
OFF-state: DSA
Pizzo et al. 2010
108
[Hz]
109
1010
Figure 5. Radio synchrotron spectra. Lines are derived from simulations, while the black crosses are compiled from observations Pizzo
(2010). The solid lines show the DSA and reaccelerated CRs (On-state of the radio halo), while the dotted lines show CRs accelerated
only by DSA (Off-state of the radio halo). The left figure shows the radio emission induced by primary CRes and the right figure shows
the emission from secondary CRes. The different line colours represent our different models, M-primaries (red line), M-streaming (blue
line), and M-turbulence (green line).
of turbulence as parametrized by tu (defined by IL thtu ,

where IL 33L kL is the injection rate of turbulence), as well
as the spatial CR profile that we will parametrize by CR,spat
(see below). We hold fixed thermal plasma properties (temperature and density profiles), B-field profiles, total CR energy content, and the turbulent outer scale (corresponding
to a wavenumber kL ). The CR energy content is suggested
by our simulations (observations only give an upper bound;
Arlen et al. 2012). We focus on the uncertain CR distribution rather than the overall normalisation, since the impact
of the latter (an overall linear scaling) is clear.
In order to quickly explore this parameter space, we
solve the Fokker-Planck equation in static spherical shells
for injection, reacceleration, and losses of the CRs, i.e., we
neglect Lagrangian evolution during re-acceleration. This
ignores the effect of adiabatic compressive heating of the
CRs, though this is generally subdominant (e.g., see Figure 7 of Miniati 2015). Once the CRs have been reaccelerated for cl = 650 Myr, we calculate the resulting radio emission numerically using the formalism outlined in
Rybicki & Lightman (1979) and compare the emission profiles and spectra as we vary one parameter at a time relative
to our fiducial model.
We adopt both the density (Briel et al. 1992) and
temperature profiles (Bonamente et al. 2009; Arnaud et al.
2001) derived from X-ray observations of the Coma cluster,
ne
kB T
h
i1.125
n0 1 + (R/Rc )2
,
h
i0.32
8.25 keV 1 + (2R/R200 )2
,
(25)
with n0 = 3.4 103 cm3 . The virial and core radii of Coma
ohringer 2002) and
are given by R200 = 2.3 Mpc (Reiprich & B
Rc = 294 kpc, respectively. In accordance with Faraday rotation measure measurements, we assume B(r) = B0 (n/n0 ) ,
where B0 = 4.8G and = 0.5 (Bonafede et al. 2010).
The bulk of the CRps are injected by relatively
low Mach number shocks and parametrized by fp,inj (p) =
Cinj pinj , where inj 2.5 in our simulations. The CRps
approximately trace the thermal gas with Cinj thCR,spat

(Pinzke & Pfrommer 2010; Vazza et al. 2016), where the
normalisation is fixed by the injection rate of CR energy in
the last 650 Myrs. Our simulations show that the CR energy
approximately amounts to 0.03 per cent of the thermal en1
R
RR
R
=
ergy inside the virial radius, i.e. 0 200 CR,inj dV 0 200 th dV
0.0003. The spectrum of the initial CRp distribution is determined by the steady state between injection and cooling,
fp,inj (p )dp
p
fp,0
,
p
dp
+
dt Coul had
where
we
fix
the
1
R
R R200
R200
dV
dV
th
CR,0
0
0
(26)
normalisation
=
by
requiring
0.003. Note that the in-
jected CR energy in the last 650 Myr is smaller by about

a factor of 10 in comparison to the cumulative CR energy
injected over the entire cosmological history of the cluster.
However, since the injected CR energy rate averaged over
the formation time of the cluster is similar to that during a
merger, the CRes injected during the merger are especially
important for the radio emission above 1 GHz since these
CRs have not yet had time to cool. Similarly, the initial
CRe distribution is given by the steady state between
cooling (Coulomb, inverse Compton, and synchrotron) and
injection of secondary CRes from fp,0 . The diffusion constant
tu 1
T kL is calculated for each radial bin. All
D pp Xtu2 th
parameters and assumptions are similar to what is used for
our simulated cluster (see section 3). Our fiducial model
assumes Xtu = 0.2, tu = 0.8, CR,spat = 1.0, and kL = 2/L
where L = 100 kpc. We also assume a fixed acceleration
time of cl = 650 Myr.
MNRAS 000, 1?? (2015)

radio spectra
radio profiles
0
34
10
1033
10-2
10
L () [erg/s/Hz]
S [Jy arcmin-2]
10
-4
Xtu = 0.01
Xtu = 0.05
Xtu = 0.1
Xtu = 0.2
Xtu = 0.25
Xtu = 0.3
352 MHz
10-6
10-8
1032
1031
1030
0.1
1029
107
1.0
Xtu = 0.01
Xtu = 0.05
Xtu = 0.1
Xtu = 0.2
Xtu = 0.25
Xtu = 0.3
Pizzo et al. 2010
108
RT / R200
0
10
tu = 0.6
tu = 0.7
tu = 0.8
tu = 0.9
tu = 1.0
352 MHz
-6
10-8
1031
1029
107
1.0
109
1010
tu = 0.6
tu = 0.7
tu = 0.8
tu = 0.9
tu = 1.0
Pizzo et al. 2010
108
RT / R200
0
[Hz]
34
10
1033
10-2
L () [erg/s/Hz]
S [Jy arcmin-2]
1010
1032
1030
0.1
10-4
10
109
1033
-4
10
1010
10
10-2
10
[Hz]
109
34
L () [erg/s/Hz]
S [Jy arcmin-2]
10
11
CR,spat = 0.0
CR,spat = 0.5
CR,spat = 0.75
CR,spat = 1.0
CR,spat = 1.25
352 MHz
-6
10-8
1032
1031
1030
0.1
1.0
RT / R200
1029
107
CR,spat = 0.0
CR,spat = 0.5
CR,spat = 0.75
CR,spat = 1.0
CR,spat = 1.25
Pizzo et al. 2010
108
[Hz]
Figure 6. Sensitivity of the radio emission in the Coma cluster to critical parameters. The left-hand panels show the radio surface brightness profiles. We compare profiles at 352 MHz (blue crosses, Brown & Rudnick 2011) to predicted emission from Fermi-II reaccelerated
CR electrons (black lines). The right-hand panels show radio synchrotron spectra. The green crosses are compiled from observations
(Pizzo 2010), while the black lines show predicted spectra. The upper panels show the sensitivity to the level of turbulence (Xtu ), the
middle panels show the impact of different turbulent profiles (tu ), and the lower panels show the dependence on spatial distributions of
initial and injected CRs (CR,spat ). We adopt the following fiducial values for our model (solid lines), Xtu = 0.2, tu = 0.8, and CR,spat = 1.0
and vary each parameter separately in each row of panels. We find that the radio emission is most sensitive to the level of turbulence.
The abundance of CR seeds, and the spatial distribution of CRs and turbulence are second-order effects.
MNRAS 000, 1?? (2015)
12
4.2

Results
Figure 6 shows the impact of turbulence and the spatial

distribution of CRs on the radio emission.
Impact of overall level of turbulence (Xtu ). From
the top panels of Figure 6, we see that as Xtu increases, there
are 3 important changes: an exponential increase in radio
luminosity, a flattening of the radio surface brightness profile, and an increase in spectral curvature. We discuss these
in turn.
As we show in Section 4.3, the exponential increase
in radio surface brightness is easily understood from equation 30, since for a power-law initial distribution function
fi (p) = C0 pinj , then Creacc C0 exp(Acl /D ) exp(B(r)Xtu2 ).
This exponential sensitivity is somewhat modified by cooling (which results in a non power-law spectrum; also, the
shorter the acceleration time, the larger the pool of seed
electrons available which would otherwise cool away), but
overall is a good approximation.
An increase in Xtu flattens the surface brightness profile,
since high acceleration efficiency leads to larger amplification
in the cluster outskirts (where cooling is less important) than
the centre. In particular, in the cluster outskirts, the reduced
impact of Coulomb cooling implies that there is a larger
pool of low-energy electrons available for reacceleration (see
timescales in Table 1).
Larger levels of turbulence also increase spectral curvature, which might seem puzzling. It can be understood
as follows. The pre-acceleration electron distribution function results from the competition between hadronic injection
f pinj and cooling, which results in a quasi-steady state
for the non-thermal (secondary) electrons. At low momenta,
when the Coulomb cooling time is short, f pinj +1 , at
high momenta, when inverse Compton/synchrotron cooling
dominates, f pinj 1 . In between, there is a quasi-adiabatic
regime where electrons accumulate (for more details, and an
analytic self-similar solution, see Sarazin 1999; Pinzke et al.
2013). In the absence of cooling, momentum advection in the
limit D pp p2 (so that D = p2 /4D pp is momentum independent) simply shifts the distribution f (p) f (Ap). When the
acceleration efficiency is low, most observable emission corresponds to the power-law tail of the distribution function
f pinj 1 set by the balance between injection and synchrotron/IC cooling. However, as the acceleration efficiency
A increases, radio emission starts to probe the bump around
p (given by D cool (p )) where electrons accumulate and
the distribution function is curved. This results in a curved
emission spectrum. The synchrotron spectrum steepens at
the frequency (Brunetti et al. 2001):
s
B2
D
(B2 + B2CMB )2
(27)
where BCMB (8CMB )1/2 , which increases for shorter D .

Impact of spatial profile of turbulence (tu ). From
the middle panels of Figure 6, we see that as expected, a flatter profile of the turbulent pressure directly translates into a
flatter radio surface brightness profile. Since seed electrons
are more concentrated toward the centre (the collisional production of secondaries is more rapid there), and magnetic
fields are stronger, concentrating the turbulence toward the
cluster centre for fixed Xtu results in higher radio luminosities, and slightly more curvature (due to the increased im-
portance of cooling near the centre). Overall, however, the

spatial profile of turbulence has a much weaker effect than
its overall normalisation.
Impact of spatial profile of seed CRs CR,spat . The
spatial distribution of CRs has an even smaller impact on
radio emission. At fixed total CR energy content XCR , concentrating the CRs towards the centre leads to more centrally dominated surface brightness profiles, as expected,
and higher radio luminosities (for the same reasons as above:
secondaries are more easily produced in the centre, and magnetic fields are stronger). We have also confirmed that radio surface brightness profiles scale linearly with XCR , as expected.
Overall, our results suggest that radio haloes are much
more sensitive to the level of turbulence (exponential dependence) rather than CR abundance (linear dependence), and
that the spatial distribution of turbulence and CRs, while
important, are second-order effects. In our parametrization,
the most important controlling variable is Xtu . The overall
level of turbulence has to be such that D cl (see Table 1), otherwise, too little or too much amplification takes
place. For instance, for Xtu & 0.08, changing Xtu by a factor
of two changes the radio surface brightness by a factor of
10 100 (see top panels of Figure 6). The required D /cl
depends only logarithmically on the abundance of seed CRs.
While the requirement of a threshold level of turbulence
may explain why radio brightness is bimodal, it also raises a
fine-tuning problem: why is the Lradio vs. LX relation in active
radio haloes so tight (Brunetti et al. 2009)? Depending on
the details of infall or mergers, we would naturally expect
fluctuations in Xtu , which would translate into large scatter
in the Lradio LX relation. This can be only be understood
if the timescale over which acceleration takes place cl also
depends on the properties of turbulence, so that the ratio
D /cl has relatively little scatter. We address this next.
4.3
Self-limiting turbulent reacceleration
In this section, we explore the physical origin of the high sensitivity of radio emission to turbulence levels (e.g., top right
panel of Fig. 6), which thus requires strong fine-tuning to
explain observed radio profiles. We will find that by relating
the acceleration time to the turbulent decay time, this sensitivity can be eliminated. We also challenge the common
assumption of assuming a Kraichnan spectrum, beginning
at the outer scale (see Section 2.2). Since turbulence is essentially hydrodynamic at large scales, this is unjustified.
Instead, we shall assume that the outer scale of the compressive fast modes is the Alfven scale.
There are 3 important timescales in this problem: the
acceleration time D , the duration over which turbulence is
active and acceleration takes place, cl , and the cooling time
cool (p). Only cool (p) is momentum dependent (and is different for ions and electrons). Thus, the outcome of acceleration depends essentially on two dimensionless numbers,
cl /D , and cool /D .
We have seen that the radio luminosity depends very
sensitively on cl /D , through the very sensitive dependence
on Xtu (Figure 6; note from equation (15) that D Xtu2 ).
Since this raises questions of fine-tuning in Xtu to explain
the observations, it is worth understanding this property in
more detail. To this end, we ignore cooling, which is a good
MNRAS 000, 1?? (2015)

approximation for the CR protons, because hadronic cooling
times are long in comparison to the other relevant timescales
of the problem.5 In this case, p = p/D , and after a time cl ,
we have
p p exp(cl /D )
(28)
(where we have used the fact that D is momentum independent). For an initial power law distribution function f (p) =
C pinj , this momentum increase can be rewritten as a change
of normalisation, f (p) = C pinj , where C = exp(inj cl /D ).
A slightly more careful derivation by direct solution of the
Fokker-Planck equation yields:
2
f
f =
p D pp
,
p
p p2
(29)
with the analytic solution given by

C = C exp
"
#
(2 + inj )(inj 1) t
.
4
D
(30)
At t = cl the CR distribution is exponentially sensitive to

cl /D , which is the physical reason underlying the extreme
sensitivity to Xtu in Figure 6.
The natural solution to this puzzle would be a process
that couples cl and D . There are two possible limits. Let us
suppose that the timescale on which there exists a source of
turbulent driving is drive , which approximately corresponds
to the timescale of the merger, or the dynamical time. Let
the turbulence decay on a timescale decay . If drive > decay ,
then cl drive , which depends on the details of the merger
and should result in considerable scatter in cl /D in different
systems. On the other hand, if decay > drive , then cl decay .
Since D and decay are both related to properties of the turbulence, it is conceivable that this would result in much less
scatter in cl /D .
An obvious candidate for decay is:
edd
L
=
3L
(31)
i.e., the eddy turnover time at the outer scale. This is subject to uncertainties about the location of the outer scale L;
estimates in the literature range from L 0.1 1 Mpc. It is
also worth remembering that MHD turbulence only applies
for l < lA , where lA is the Alfven scale where 3 3A . Invoking
fast modes, Kraichnan scalings, etc., is only valid below these
scales. For l > lA , turbulence is basically hydrodynamic.
In the hydrodynamic regime, a standard HodgeHelmholtz decomposition usually shows that the compressive component of the velocity field is Burgers-like (W(k)
k2 ), while the solenoidal component is Kolmogorov-like
(W(k) k5/3 ), see e.g., Federrath (2013). The Burgerslike component does not reflect a genuine cascade, but
rather the appearance of shocks which directly transfer
power from large to small scales. At face value, we should
use the Burgers spectrum for compressible modes. However, as already found by previous authors (Miniati 2015;
Brunetti & Lazarian 2016), and as we shall discuss, this
does not produce significant particle acceleration. If this is
5
In principle, CRp can still lose energy via wave heating, at a

rate |3A PCR |, but we assume CR streaming is suppressed during
mergers, when they are spatially confined by scattering.
MNRAS 000, 1?? (2015)
13
the correct spectrum, then the paradigm of turbulent particle acceleration is simply flawed, and some other mechanism is necessary to explain radio halos. Alternatively, it
is well-known that due to mode-mode coupling, solenoidal
modes can give rise to compressive modes, and vice-versa
(Kida & Orszag 1990; Cho & Lazarian 2003; Kritsuk et al.
2007), even for subsonic turbulence, since pressure fluctuations of order u2 arise. Indeed, note that many numerical studies (such as Cho & Lazarian (2003); Kritsuk et al.
(2007)) only have solenoidal driving on the outer scale, but
then are also able to study the compressive modes that
develop. In hydrodynamic turbulence, the energy in compressible modes which develop in this way scales as M2s ,
for M s < 1; this coupling is strongest at the Alfven scale
(Cho & Lazarian 2003). In MHD turbulence, mode-mode
coupling is weak below the Alfven scale and the Alfven, fast
and slow modes proceed as separate cascades.
We thus assume that some fraction of the Kolmogorovlike hydrodynamic turbulence which cascades down to
the Alfven scale ends up as compressive fast modes
with a Kraichnan spectrum, as seen in simulations
(Cho & Lazarian 2003). Henceforth, we can consider the
outer scale of the fast modes to be lA . If some fraction fc
of the turbulent energy density at this scale is in compressible modes, then the energy density at the outer scale is
fc 32A . The turbulent reacceleration time is:
D =
p2
CD c
= 1/2
4D pp
A kA fc 32A
(32)
where CD = 2/(5), A 11000, and we have used equations, (6), (7), (14), assuming s = 3/2. The turbulent decay
time is given by the cascade time of fast modes at the Alfven
scale k = kA (Yan & Lazarian 2004; see equation 11):
decay =
3ph
cs
=
.
32k k
fc 32A kA
(33)
This can be related to the eddy turnover time at the outer

scale edd from lA = LM3
A to:
decay =
L c2s 3A
X tu
1.0 edd
2
3L fc 3L cs
0.2
!1
1/2
50
(34)
where we have used X tu fc (3L /cs )2 and cs /3A = 1/2 . The

decay time at the Alfven scale is comparable (and could be
larger, if fc is smaller than we have assumed) to the eddy
turnover time at the outer scale. Thus, it is appropriate to
consider the latter as the decay time for the fast modes,
rather than the cascade time at the driving scale. The reason
for these comparable timescales despite the disparate length
scales is that large scale hydrodynamic modes cascade on a
single eddy turnover time, whereas small scale sub-Alfvenic
MHD modes are in the weak turbulence limit, and require
multiple wave-wave interactions for a mode to cascade.
From equations (32) and (33), we obtain:

1
decay
cl
A1/2 cs 1
cs
.
=
= 0.1
D
D
CD c
1500 km s1 50
(35)
Remarkably, this expression is independent of properties of

the turbulence such as 3c , L, or lA , and depends only on properties of the plasma (cs , ). Ultimately, this arises because the
timescale on which a fast mode wave transfers energy due
14
to wave particle interactions in second order Fermi acceleration, p [4 kp c(3c /c)2 ]1 , is closely related to the timescale
on which it cascades due to wave-wave interactions, w
cs /(kw 32c ). This implies w /p 4 (cs /c)(lw /lp ), where lw lA
is the outer scale on which the fast mode cascade begins,
and lp (lA lcut )1/2 is the characteristic wavelength for waveparticle interactions. For transit time damping and an outer
scale of lA , we have lcut (me /mp )lA (cs /3A )4 . We thus have
w /p 4 (cs /c)(lA /lcut )1/2 4 (cs /c)(mp /me )1/2 1 . More careful
consideration of dimensionless factors boosts this estimate
by an order of magnitude to give equation (35).
On the other hand, equation (35) points toward a pessimistic scenario where turbulent reacceleration with TTD
on thermal particles is never effective. A key reason is that
we assume Kraichnan turbulence only applies for l < lA .
There is then insufficient separation of scales: the cutoff
scale lcut 0.2250 lA . Although there is a fair large separation of scales between the outer driving scale L and
lA = LM3
A (a factor of 30 1000 for MA 3 10; we
have MA (X tu )1/2 3.2 for our fiducial assumptions), Kolmogorov turbulence, with its steeper spectrum, has more
energy at large scales (kW(k) k2/3 , k1/2 for Kolmogorov
and Kraichnan turbulence, respectively). This implies that
the energy-weighted scales at which wave-particle interaction take place are larger in Kolmogorov turbulence, and
thus that the wave-particle interaction rate is lower. If (as is
frequently seen) we instead assume that Kraichnan turbulence begins at the outer scale L, with characteristic decay
time Lcs /32c , then we obtain a more palatable result:

X !
cl
cs
tu
0.8
.
(36)
D
1500 km s1 0.2
This arises because the decay time is now MA times longer,

and the acceleration time is now M1/2
A times shorter, boosting cl /D by M3/2
A . However, as we have argued, turbulence
in the hydrodynamic regime lA < l < L is Kolmogorov, not
Kraichnan. Furthermore, if this scaling is correct, then this
leaves us with the problematic exponential sensitivity to Xtu
that we previously explored.
One alternative is that scattering in the high- ICM
is mediated by plasma instabilities (firehose, mirror) rather
than Coulomb scattering. This vastly increases the scattering rate and reduces the mean free path of thermal particles. In this case, the fast modes damp by TTD on relativistic rather than thermal particles. The momentum diffusion
coefficient in this case is then (Brunetti & Lazarian 2011;
Miniati 2015):
DCR
pp =
2p2 h32c i2
kL 3
XCR
cs
(37)
where is an efficiency factor for the effectiveness of plasma

instabilities (e.g., due to spatial or temporal intermittency),
and XCR = CR /th is the relative energy density of CRs. If we
set kL kA , 32c 32A , as before, then:
cl
8
D
XCR
(Brunetti & Lazarian 2011). One natural assumption is to

assume that CR saturates at a level CR turb (MHD) B ,
in which case XCR 1 , so that:

cl
2.5
.
(39)
D
0.3
Given the host of uncertain factors which enter into < 1

(the intermittent nature of turbulence, small scale magnetic
topology and the efficiency with which instabilities mediated
by pressure anisotropies are triggered), this is potentially
consistent with cl D . This promising scenario should be
investigated in more detail.
It is also worth exploring the effect of Burgers turbulence (weak shocks). Burgers turbulence is distinct from
other forms of turbulence in that it violates locality in kspacei.e., power does not gradually cascade, but instead
can directly jump from large to small scales via a network
of weak shocks. This implies a short dissipation time and we
have in this case, cl drive , the driving time, rather than the
turbulent dissipation time. Since the Fourier transform of a
step function goes as k1 , Burgers turbulence has a power
spectrum of P(k) k2 . From equations (6) and (7), we find:
!1/2
1/2 X 1/2
DBurgers
pp
tu
(40)
=
=
0.15
DMHD
Xtu A
50
0.2
pp
where we have set hkiW kL for Burgers turbulence, i.e.
all power is at large scales. This weighting towards larger
scales implies a much lower wave-particle interaction rate
and thus a diffusion coefficient which is smaller than standard TTD on thermal particles by an order of magnitude.
Since hkiW kL is independent of the cutoff scale kcut , this
conclusion is unchanged if plasma instabilities regulate the
thermal particle mean free path and TTD operates on relativistic particles instead. Since standard TTD on thermal
particles was already potentially problematic (equation 35),
we conclude, in agreement with other assessments (Miniati
2015; Brunetti 2016), that if Burgers turbulence dominates,
then turbulent reacceleration will be ineffective.
It is interesting to reconsider surface brightness and
spectral profiles if indeed D cl for the reasons mentioned
above (e.g., equation 39). We show the results of adopting
such an ansatz in Figure 7, where, similar to Figure 6, we
vary Xtu , tu , and CR,spat about fiducial values.6 Instead of
adopting a fixed cl , we set cl /D = 2.5. The results are
remarkably revealing. In this case, we are indeed less sensitive to the properties of the turbulence, and more sensitive to the details of the CR seed population. This gives
hope to the possibility that one could effectively marginalise
over the very uncertain properties of turbulence at larger
scales in merging clusters (which are unlikely to be more
precisely constrained observationally in the near future) to
learn something about the underlying CR population.
For instance, in the third row of panels of Figure 7, we
see that once turbulence exceeds a threshold value Xtu 0.2,
(38)
In principle, this implies exponential sensitivity to XCR , a

similar situation as exponential sensitivity to Xtu . However,
there is a self-limiting asymptotic behaviour: as XCR increases due to turbulent reacceleration, damping increases,
which limits the further growth of the CR energy density
When cl /decay const is adopted, the radial behaviour of cl is

also modified. Because D decreases with radius (see e.g. Figure 4),
cl has to decrease by the same amount, which results in a steeper
radio profile. To compensate for this effect in Figure 7, we change
CR,spat = 1.0 0.5 and decrease the injection rate of CR energy
by a factor five.
MNRAS 000, 1?? (2015)

then its overall exact energy density does not matter profiles simply converge to an asymptotic form as Xtu is increased. This is in contrast to the exponential sensitivity,
flattening in surface brightness, and increasing spectral curvature as Xtu is increased that we saw previously. A threshold value of Xtu is necessary for turbulent reacceleration to
overcome cooling. Once that condition is satisfied, the condition cl /D const implies a fixed amount of amplification:
stronger turbulence implies faster acceleration, but also dissipates more quickly. This leads to the asymptotic behaviour
seen. The value of this constant depends on plasma physics
details, but we have shown that the value required to match
observations is plausible.
The second row of panels of Figure 7 reveal another
striking result: in contrast to Figure 6, results are also completely insensitive to the spatial profile of turbulence. Given
the large uncertainties, this is a welcome property. It arises
because for our fiducial value (Xtu = 0.2), variations in the
turbulent profile still lead to local energy densities which are
above the threshold value required to overcome cooling, and
turbulent amplification approaches its asymptotic value.
Finally, in the top panels of Figure 7, we see that we
are sensitive to the spatial profile of CRs; it affects both the
radio surface brightness profile (a flatter CR distribution implies flatter surface brightness profiles) and luminosity (once
again, luminosity is larger for more centrally concentrated
CRs, since more secondaries are produced). This arises because once turbulent reacceleration overcomes cooling, the
condition cl /D const provides a fixed amount of amplification in each radial shell. The radio luminosity in each shell
then depends linearly on initial conditions, i.e., the profile
of CR seeds. Interestingly, the best-fitting profile for Coma
results from a flat CR distribution i.e., the outcome of efficient CR streaming. The overall amplitude is set by cl /D
(equation 39) and the CR content.
Obviously, the ideas in this section require further
study. For instance, the notion that fast modes appear
abruptly at lA and that they last for a time cl are clear approximations (the latter could be improved by solving the
time dependent diffusion equation for the fast mode energy
spectrum, see e.g. Zhou & Matthaeus (1990)). However, the
notion that cl /D could self-regulate around a fixed value
is exciting, because it eliminates the main source of uncertainty the poorly unconstrained properties of turbulence,
and implies that radio haloes could potentially provide more
robust constraints on the underlying CR population.
CONCLUSIONS
The standard reacceleration model for radio haloes requires

two ingredients with no current direct observational constraints: a CR population which produces seed electrons,
and turbulence to perform second-order Fermi acceleration
on these electrons7 . For the best studied radio halo, Coma,
there are two main observational constraints: the radio surface brightness profile S (r), and the integrated luminosity as
a function of frequency L(). For certain assumptions about
7
Other model ingredients, such as temperature, density, and Bfield profiles, are observationally constrained by X-ray and Faraday rotation measurements.
MNRAS 000, 1?? (2015)
15
turbulence and the seed population, analytic theoretical

models can match these observations (Brunetti & Lazarian
2011), but the realism of the assumed seed electron and turbulent profiles must be confronted with numerical simulations.
Cluster turbulence has been studied numerically in cosmological simulations with an eye toward radio halo properties (Beresnyak et al. 2013; Miniati 2015), though they
were not confronted against the two benchmark observations mentioned above. Most importantly, direct simulation
of the seed electron population has been missing in the literature. In this paper, we fill that gap. At the same time,
we study the sensitivity of radio halo profiles and spectra to
variation in properties of turbulence and seed population,
and uncover the dominant effects.
In the first part of the paper, we use hydrodynamic
zoom simulations of a Coma-like cluster in a cosmological setting to generate spatially and momentum-resolved
seed populations of CRps and CRes from diffusive shock
acceleration. The simulations include time-dependent cooling and adiabatic transport processes, and has been previously used in a variety of applications in CR cluster physics
(Pinzke & Pfrommer 2010; Pinzke et al. 2013). The resultant CRp and CRe distribution functions then serve as initial conditions for our calculations of Fermi II acceleration
by turbulence. We use time-dependent Fokker-Planck calculations which allow for spatial variations in the properties
of turbulence (i.e., a spatially varying momentum diffusion
coefficient). We adopt a simple parametric model for cluster turbulence, characterising it by its overall normalisation,
spatial profile, and characteristic lifetime cl . Our approach
is relatively sophisticated in its treatment of the CR seed
population and relatively simplified in its treatment of the
turbulence; it is thus orthogonal and complementary to existing treatments which focus on the time-dependent compressible turbulence (Miniati 2015).
Given our seed electron population (which includes both
primaries generated by DSA and secondary electrons created
by CRps in hadronic interactions; for standard assumptions,
the latter dominates), we find that applying the standard
model previously used to reproduce Comas radio halo properties (Brunetti & Lazarian 2011), in which turb th , fails:
it produces radio emission which is too centrally concentrated. Indeed, Brunetti & Lazarian (2011) remark in their
paper that the CR population required in their model is
remarkably flat. By contrast, the simulations produce CRs
which are more centrally concentrated.
Thus, some modification of the standard model is
needed, either by modifying the spatial distribution of the
CRs, or that of turbulence. We explore two examples of the
former: (i) CRs can stream in the ICM, producing a flat profile (Enlin et al. 2011; Wiener et al. 2013), or (ii) given a
higher e/p ratio in DSA acceleration (perhaps due to magnetic geometry; Guo et al. 2014a), the primary population
can dominate. This has a much flatter profile (since secondaries must be generated collisionally, they will always
be concentrated towards denser regions). However, both of
these solutions still require some modification of the turbulent profile, requiring turb /th to rise with radius. Indeed,
one can simply use the unmodified CR profile derived from
simulations if (iii) turb /th rises somewhat more steeply with
radius. The fact that turb /th rises with radius is well sup-
16

radio spectra
radio profiles
100
1034
Xtu = 0.01
Xtu = 0.05
Xtu = 0.1
Xtu = 0.2
Xtu = 0.25
Xtu = 0.3
Pizzo et al. 2010
10
10-2
L () [erg/s/Hz]
S [Jy arcmin-2]
33
-4
10
Xtu = 0.01
Xtu = 0.05
Xtu = 0.1
Xtu = 0.2
Xtu = 0.25
Xtu = 0.3
10-6
1031
1030
352 MHz
100
1034
tu = 0.6
tu = 0.7
tu = 0.8
tu = 0.9
tu = 1.0
Pizzo et al. 2010
1033
10-2
L () [erg/s/Hz]
S [Jy arcmin-2]
1032
-4
10
tu = 0.6
tu = 0.7
tu = 0.8
tu = 0.9
tu = 1.0
10-6
1032
1031
1030
352 MHz
100
1034
CR,spat = 0.0
CR,spat = 0.5
CR,spat = 0.75
CR,spat = 1.0
CR,spat = 1.25
Pizzo et al. 2010
10
10-2
L () [erg/s/Hz]
S [Jy arcmin-2]
33
-4
10
CR,spat = 0.0
CR,spat = 0.5
CR,spat = 0.75
CR,spat = 1.0
CR,spat = 1.25
10-6
1032
1031
1030
352 MHz
100
1034
cl/D = 1.0
cl/D = 1.5
cl/D = 2.0
cl/D = 2.5
cl/D = 3.0
Pizzo et al. 2010
10
10-2
L () [erg/s/Hz]
S [Jy arcmin-2]
33
-4
10
cl/D = 1.0
cl/D = 1.5
cl/D = 2.0
cl/D = 2.5
cl/D = 3.0
10-6
1032
1031
1030
352 MHz
0.1
1.0
RT / R200
107
108
[Hz]
109
1010
Figure 7. Sensitivity of radio emission in the Coma cluster to critical parameters when we vary the ratio of the time scales on which
turbulence is active (cl ) to the turbulent re-acceleration time (D ). In comparison to Figure 6, we change our fiducial model (solid lines)
and adopt a slightly more extended spatial distribution of CRs (CR,spat = 0.5) and lower the injection rate of CR energy by a factor five;
otherwise, we adopt Xtu = 0.2, tu = 0.8, cl = 2.5 D as fiducial values and vary each parameter separately in each row of panels. Radio
surface brightness profiles (left) are contrasted to radio synchrotron spectra (right; for more details see caption to Figure 6). Provided
there is a minimum (threshold) level of turbulence (Xtu & 0.2), here the radio profiles are mainly determined by the CR distribution,
while turbulence has a remarkably small impact. In the bottom two panels we vary the exponential momentum growth factor cl /D (see
equations 28 and 30), which mainly changes the normalisation of the CR distribution and neither spectral shape nor radial synchrotron
profile. Interestingly, the flat CR profiles adopted here are similar to those required to reproduce the radio profiles in our CR streaming
model (see Figure 2).
MNRAS 000, 1?? (2015)

ported by cosmological hydrodynamic simulations of clusters
(Lau et al. 2009; Shaw et al. 2010; Vazza et al. 2011a); even
the last model (iii) is (within the scatter) perfectly consistent with the trends seen by such simulations. This strongly
suggests that a rising turb /th should be part of any final
model of radio haloes. Given the lack of observational constraints on turbulent profiles, this unfortunately means that
models will have yet another degree of freedom.
The above models are merely meant to serve as existence proofs, showing what modifications to the standard
model are necessary to be reconciled with the radio halo observations. There are too many uncertainties and parameter
degeneracies for any of these models to be definitive, or for
one to make firm quantitative statements about turbulent
and seed CR profiles, apart from placing certain bounds.
We abstract the main parameter dependencies by exploring
static, spherically symmetric models consistent with Coma,
where we repeat our Fokker-Planck calculations. We explore
three main effects: the overall normalisation of turbulence,
and spatial distribution of turbulence and CRs.
We find that the most important variable is the overall
normalisation of turbulence: surface brightness scales exponentially with the amount of turbulence. The radio spectrum also becomes more curved with higher turbulence. In
contrast, the surface brightness scales linearly with the CR
abundance. The spatial distributions of both turbulence and
CRs also influence radio halo profiles (for both, flatter distributions at fixed overall normalisation imply flatter surface
brightness profiles with lower overall radio luminosity), but
play a secondary role.
This exponential dependence of radio halo luminosity with turbulence then raises the interesting question of
why radio halo scaling relations (e.g., L LX ) are so tight.
Previous statistical modelling of radio halo populations
(Cassano et al. 2006, 2007) have focused on matching the
mean relations, but not the scatter. We believe the latter
provides a particularly interesting and potentially fruitful
constraint. The acceleration timescale D and the length
of time during which acceleration takes place cl must be
matched so that there are roughly 2 e-folds (given the
factor 10 difference in radio luminosity between the radio bright and faint populations; Brown & Rudnick (2011)),
otherwise scatter in turbulence (which we might expect
given the wide variety of merger and infall conditions) will be
exponentially amplified in radio luminosity. The timescale of
the merger has some natural scatter which is not obviously
correlated with D .
If instead we adopt cl /decay const, i.e. the natural
lifetime of turbulence is its decay lifetime, then we can
show that the acceleration time and the lifetime of turbulence are coupled, simply because the wave-particle interaction rate (which drives particle acceleration) and the wavewave interaction rate (which drives dissipation) are linked.
If we are mindful that Kraichnan turbulence is only applicable below the Alfven scale lA (where turbulent velocities 3 < 3A ), then remarkably the ratio cl /D is independent of properties of the turbulence (such as its amplitude
and outer scale) and only depends on plasma parameters.
Our results suggest that TTD on thermal particles may result in overly rapid damping of turbulence, and inefficient
acceleration (equation 35), but if plasma instabilities scatter the thermal particles so that the fast modes damp via
MNRAS 000, 1?? (2015)
17
TTD on the CRs, then the required acceleration can take

place (equation 39). In agreement with other assessments,
we find that if turbulence is Burgers rather than Kraichnan, then turbulent reacceleration is ineffective, and some
other explanation for radio haloes (e.g., magnetic reconnection, Brunetti & Lazarian 2016) is necessary. If we adopt the
ansatz cl /decay const, then we find that above a threshold
level of turbulence (necessary to overcome cooling), radio
properties are insensitive to both the amplitude and spatial
profile of turbulence, since only a fixed, asymptotic amount
of amplification takes place. On the other hand, they are
sensitive to the properties of the seed electron profile. This
raises the exciting possibility that one could marginalise over
the highly uncertain properties of cluster turbulence to learn
about the underlying seed CR population. Our results suggest that studying not just mean trends, but the relatively
small scatter in radio-halo scaling relations, may prove extremely fruitful in shedding light on ICM plasma processes.
Acknowledgements. We thank Josh Wiener for discussions on CR streaming, Lawrence Rudnick for discussion on uncertainties in the 1.4 GHz radio data, Fabio
Zandanel for re-calculating gamma-ray limits, and Gianfranco Brunetti for useful discussions. We thank an anonymous referee for helpful comments that improved the paper.
A.P. is grateful to the Swedish research council for financial
support. S.P.O. thanks NASA grants NNX12AG73G and
NNX15AK81G for support, as well as KITP for hospitality.
This research was supported in part by the National Science
Foundation under Grant No. NSF PHY11-25915. C.P. acknowledges support by the European Research Council under ERC-CoG grant CRAGSMAN-646955 and by the Klaus
Tschira Foundation.
REFERENCES
Ackermann M., et al., 2014, ApJ, 787, 18
Ahnen M. L., et al., 2016, A&A, 589, A33
Arlen T., et al., 2012, ApJ, 757, 123
Arnaud M., et al., 2001, A&A, 365, L67
Basu K., 2012, MNRAS, 421, L112
Beresnyak A., Xu H., Li H., Schlickeiser R., 2013, ApJ, 771, 131
Blasi P., Colafrancesco S., 1999, Astroparticle Physics, 12, 169
Bonafede A., Feretti L., Murgia M., Govoni F., Giovannini G.,
Dallacasa D., Dolag K., Taylor G. B., 2010, A&A, 513, A30
Bonamente M., Lieu R., Bulbul E., 2009, ApJ, 696, 1886
Briel U. G., Henry J. P., Boehringer H., 1992, A&A, 259, L31
Brown S., Rudnick L., 2011, MNRAS, 412, 2
Brunetti G., 2016, Plasma Physics and Controlled Fusion,
58, 014011
Brunetti
G.,
Jones
T.
W.,
2014,
International Journal of Modern Physics D, 23, 30007
Brunetti G., Lazarian A., 2007, MNRAS, 378, 245
Brunetti G., Setti G., Feretti L., Giovannini G., 2001, MNRAS,
320, 365
Brunetti G., Blasi P., Cassano R., Gabici S., 2004, MNRAS,
350, 1174
Brunetti G., Cassano R., Dolag K., Setti G., 2009, A&A, 507, 661
Brunetti G., Blasi P., Reimer O., Rudnick L., Bonafede A., Brown
S., 2012, MNRAS, 426, 956
Caprioli D., Spitkovsky A., 2014, ApJ, 783, 91
Cassano R., Brunetti G., Setti G., 2006, MNRAS, 369, 1577
Cassano R., Brunetti G., Setti G., Govoni F., Dolag K., 2007,
MNRAS, 378, 1565
Chernyshov D. O., Dogiel V. A., Ko C. M., 2012, ApJ, 759, 113
18
Cho J., Lazarian A., 2003, MNRAS, 345, 325

Deiss B. M., Reich W., Lesch H., Wielebinski R., 1997, A&A,
321, 55
Dennison B., 1980, ApJ, 239, L93
Dermer C. D., 1986, ApJ, 307, 47
Dolag K., Borgani S., Murante G., Springel V., 2009, MNRAS,
399, 497
Donnert J., Brunetti G., 2014, MNRAS, 443, 3564
Donnert J., Dolag K., Brunetti G., Cassano R., 2013, MNRAS,
429, 3564
Ellison D. C., Berezhko E. G., Baring M. G., 2000, ApJ, 540, 292
Enlin T., Pfrommer C., Miniati F., Subramanian K., 2011, A&A,
527, A99+
Farmer A. J., Goldreich P., 2004, ApJ, 604, 671
Federrath C., 2013, MNRAS, 436, 1245
Feretti L., Giovannini G., Govoni F., Murgia M., 2012, A&ARv,
20, 54
Giovannini G., Feretti L., Venturi T., Kim K. T., Kronberg P. P.,
1993, ApJ, 406, 399
Gould R. J., 1972, Physica, 60, 145
Govoni F., Enlin T. A., Feretti L., Giovannini G., 2001, A&A,
369, 441
Guo X., Sironi L., Narayan R., 2014a, ApJ, 794, 153
Guo X., Sironi L., Narayan R., 2014b, ApJ, 797, 47
Kang H., Ryu D., 2011, ApJ, 734, 18
Kida S., Orszag S. A., 1990, Journal of Scientific Computing, 5, 1
Kowal G., Lazarian A., 2010, ApJ, 720, 742
Kritsuk A. G., Norman M. L., Padoan P., Wagner R., 2007, ApJ,
665, 416
Lau E. T., Kravtsov A. V., Nagai D., 2009, ApJ, 705, 1129
Lynn J. W., Quataert E., Chandran B. D. G., Parrish I. J., 2014,
ApJ, 791, 71
Miniati F., 2015, ApJ, 800, 60
Miniati F., Ryu D., Kang H., Jones T. W., 2001a, ApJ, 559, 59
Miniati F., Jones T. W., Kang H., Ryu D., 2001b, ApJ, 562, 233
Park
J.,
Caprioli
D.,
Spitkovsky
A.,
2015,
Physical Review Letters, 114, 085003
Petrosian V., East W. E., 2008, ApJ, 682, 175
Pfrommer C., 2008, MNRAS, 385, 1242
Pfrommer C., Enlin T. A., 2004, A&A, 413, 17
Pfrommer C., Springel V., Enlin T. A., Jubelgas M., 2006,
MNRAS, 367, 113
Pfrommer C., Enlin T. A., Springel V., Jubelgas M., Dolag K.,
2007, MNRAS, 378, 385
Pfrommer C., Enlin T. A., Springel V., 2008, MNRAS, 385, 1211
Pinzke A., Pfrommer C., 2010, MNRAS, 409, 449
Pinzke A., Oh S. P., Pfrommer C., 2013, MNRAS, 435, 1061
Pizzo R. F., 2010, PhD thesis, University of Groningen
Planck Collaboration et al., 2013, A&A, 554, A140
Porter D. H., Jones T. W., Ryu D., 2015, ApJ, 810, 93
Quilis V., Ibanez J. M. A., Saez D., 1998, ApJ, 502, 518
Reiprich T. H., B
ohringer H., 2002, ApJ, 567, 716
Rybicki G. B., Lightman A. P., 1979, Radiative processes in astrophysics. New York, Wiley-Interscience
Sarazin C. L., 1999, ApJ, 520, 529
Sarkar S., Erlebacher G., Hussaini M. Y., Kreiss H. O., 1991,
Journal of Fluid Mechanics, 227, 473
Schlickeiser R., Sievers A., Thiemann H., 1987, A&A, 182, 21
Shaw L. D., Nagai D., Bhattacharya S., Lau E. T., 2010, ApJ,
725, 1452
Subramanian K., Shukurov A., Haugen N. E. L., 2006, MNRAS,
366, 1437
Vazza F., Br
uggen M., 2014, MNRAS, 437, 2291
Vazza F., Dolag K., Ryu D., Brunetti G., Gheller C., Kang H.,
Pfrommer C., 2011a, MNRAS, 418, 960
Vazza F., Brunetti G., Gheller C., Brunino R., Br
uggen M.,
2011b, A&A, 529, A17
Vazza F., Gheller C., Br
uggen M., 2014, MNRAS, 439, 2662
Vazza F., Br
uggen M., Wittor D., Gheller C., Eckert D., Stubbe
M., 2016, MNRAS, 459, 70
Wiener J., Oh S. P., Guo F., 2013, MNRAS, 434, 2209
Yan H., Lazarian A., 2002, Physical Review Letters, 89, 1102
Yan H., Lazarian A., 2004, ApJ, 614, 757
Zandanel F., Ando S., 2014, MNRAS, 440, 663
Zandanel F., Pfrommer C., Prada F., 2014, MNRAS, 438, 124
Zhou Y., Matthaeus W. H., 1990, J. Geophys. Res., 95, 14881
MNRAS 000, 1?? (2015)

Turbulence and Particle Acceleration in Giant Radio Haloes: The Origin of Seed Electrons

Uploaded by

Copyright:

Available Formats

Turbulence and Particle Acceleration in Giant Radio Haloes: The Origin of Seed Electrons

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Turbulence and Particle Acceleration in Giant Radio Haloes: The Origin of Seed Electrons

Uploaded by

Copyright:

Available Formats

MNRAS 000, 1??

Preprint 24 November 2016

Compiled using MNRAS LATEX style file v3.0

Turbulence and Particle Acceleration in Giant Radio

arXiv:1611.07533v1 [astro-ph.HE] 22 Nov 2016

About one third of X-ray-luminous clusters show smooth,

apinzke@fysik.su.se (AP); peng@physics.ucsb.edu

A. Pinzke, S. Peng Oh and C. Pfrommer

Figure 1. Spatial distribution of CRp energy density in the Coma

Origin of Seed Electrons

COSMIC RAY TRANSPORT

The transport of relativistic electrons and protons across

Here fe is the one-dimensional distribution in position x

1979) losses, respectively,

A. Pinzke, S. Peng Oh and C. Pfrommer

matically quiescent. Furthermore, even Alfvenic streaming

where we use the cross-section, +/,0 (p), given by the fitting

In turbulent reacceleration, particles gain energy via Fermi

Equations (6) and (8) make the important aspects of

where IL is the volumetric injection rate of turbulence at

It is important to realise that this Kraichnan scaling

Origin of Seed Electrons

where A 11000. If there is an unimpeded cascade from the

What is a typical energy density in compressive fast

Note that this differs from previous work, which assumes

In this section, we solve the Fokker-Planck equation for CR

A. Pinzke, S. Peng Oh and C. Pfrommer

Table 1. Thermal quantities and timescales for different spatial

find the minimal modifications required to fit observations.

Modelling diffusive shock acceleration

In this paper we focus on our simulated cluster, g72a, which

inj that depends on the adiabatic index ad = 5/3 and the

determined by the normalisation C(t) and the spectral index

The particle injection efficiency is a strong function of xinj

Origin of Seed Electrons

where the new distribution function is given by fp (p, t) =

Radio emission profile

In Figure 2, we show radial profiles for the radio emission

of D pp (R) th (R)tu 1 T (R). This signals that the problem

Lorentz factor on emission frequency ( syn ), this effect

Note that in previous work on the Coma cluster, turb

A. Pinzke, S. Peng Oh and C. Pfrommer

10-1 secondary CRe

10-1 primary CRe

Brunetti et al. (2012):

10-1 secondary CRe Simulations; DSA+FII reacc.

10-1 secondary CRe

In Figure 5 we show that our three models that include

clusters in the OFF-state (DSA only) and ON-state (DSA

turb(R) / th(R), Eturb(< R) / Eth(< R)

Origin of Seed Electrons

The gamma-ray emission from CRps that produce decaying

Figure 4. Turbulent reacceleration timescales for our simulated

mentioned in section 4, combining radio observations with

PARAMETER SPACE EXPLORATION AND

As we have seen before, the most uncertain aspects of radio

A. Pinzke, S. Peng Oh and C. Pfrommer

1034 Secondary CRes