Draft version February 18, 2013

Preprint typeset using LATEX style emulateapj v. 04/20/08

BUOYANCY INSTABILITIES IN GALAXY CLUSTERS: CONVECTION DUE TO ADIABATIC COSMIC RAYS

AND ANISOTROPIC THERMAL CONDUCTION
Prateek Sharma1
Theoretical Astrophysics Center and Astronomy Department, University of California, Berkeley, CA 94720

arXiv:0901.4786v1 [astro-ph.HE] 30 Jan 2009

Benjamin D. G. Chandran
Space Science Center and Department of Physics, University of New Hampshire, Durham, NH 03824

Eliot Quataert, Ian J. Parrish1

Theoretical Astrophysics Center and Astronomy Department, University of California, Berkeley, CA 94720
Draft version February 18, 2013

ABSTRACT
Using a linear stability analysis and two and three-dimensional nonlinear simulations, we study
the physics of buoyancy instabilities in a combined thermal and relativistic (cosmic ray) plasma,
motivated by the application to clusters of galaxies. We argue that cosmic ray diffusion is likely to
be slow compared to the buoyancy time on large length scales, so that cosmic rays are effectively
adiabatic. If the cosmic ray pressure pcr is & 25% of the thermal pressure, and the cosmic ray
entropy (pcr /4/3 ; is the thermal plasma density) decreases outwards, cosmic rays drive an adiabatic
convective instability analogous to Schwarzschild convection in stars. Global simulations of galaxy
cluster cores show that this instability saturates by reducing the cosmic ray entropy gradient and
driving efficient convection and turbulent mixing. At larger radii in cluster cores, the thermal plasma
is unstable to the heat flux-driven buoyancy instability (HBI), a convective instability generated
by anisotropic thermal conduction and a background conductive heat flux. The HBI saturates by
rearranging the magnetic field lines to become largely perpendicular to the local gravitational field;
the resulting turbulence also primarily mixes plasma in the perpendicular plane. Cosmic-ray driven
convection and the HBI may contribute to redistributing metals produced by Type 1a supernovae in
clusters. Our calculations demonstrate that adiabatic simulations of galaxy clusters can artificially
suppress the mixing of thermal and relativistic plasma; anisotropic thermal conduction allows more
efficient mixing, which may contribute to cosmic rays being distributed throughout the cluster volume.
Subject headings: convection cooling flows galaxies: active galaxies: clusters: general
magnetic fields
1. INTRODUCTION

Microscopic transport of heat and momentum in dilute plasmas, like those in clusters of galaxies, is primarily along magnetic field lines (Braginskii 1965). This
anisotropic transport dramatically affects the convective
stability of the plasma; convective stability is no longer
determined by the entropy gradient (Schwarzschild
1958). Instead, a plasma is unstable to buoyant motions irrespective of the background entropy and temperature gradients (Balbus 2000; Quataert 2008). Cosmic
rays diffusing along magnetic field lines also affect the
convective stability of the plasma (Chandran & Dennis
2006; Dennis & Chandran 2008). Nonlinear simulations
show that these instabilities driven by anisotropic thermal and cosmic ray transport can change the magnetic
field configuration, and the background temperature and
density profiles in the plasma, but they do not drive efficient convection (e.g., Parrish & Stone 2007; Parrish
& Quataert 2008; Parrish, Stone, & Lemaster 2008;
Electronic address: psharma@astro.berkeley.edu
Electronic address: benjamin.chandran@unh.edu
Electronic
address:
eliot@astro.berkeley.edu,
rish@astro.berkeley.edu
1 Chandra Fellow

ipar-

Sharma, Quataert, & Stone 2008). The instabilities saturate largely by rearranging the magnetic field configuration, thereby slowing down the instability and reaching
a state of marginal stability to linear perturbations. By
contrast, hydrodynamic convection in stellar interiors redistributes energy efficiently to make the plasma nearly
adiabatic and thus marginally stable to convection.
One of the key astrophysical motivations for studying
the transport properties of dilute plasmas in the presence of cosmic rays is to understand the dynamical and
thermal structure of clusters of galaxies. The radiative
cooling time (. 1 Gyr) is much less than the Hubble
time (tH 13.7 Gyr) in cluster cores. Thus, it was expected that the intracluster medium (ICM) would cool
rapidly, resulting in large rates (& 100M yr1 ) of mass
cooling to form cold gas and stars (e.g., Fabian 1994).
However, X-ray observations have failed to detect copious emission from the expected cold plasma component
in cluster cores (e.g., Peterson et al. 2003). The lack
of cooling flows implies that cooling is balanced by some
source of heating, e.g., heating by thermal conduction
from large radii (Bertschinger & Meiksin 1986; Zakamska
& Narayan 2003), heating by jets and bubbles blown by
a central AGN (Binney & Tabor 1995; Ciotti & Ostriker

2
2001), or heating by cosmic rays (e.g., Rosner & Tucker
1989; Loewenstein, Zweibel, & Begelman 1991; Chandran & Rasera 2007). Although thermal conduction
may operate at large radii, it appears that the plasma at
small radii must be heated by a feedback process which
efficiently self-regulates. This is required to avoid the fine
tuning of thermal conductivity required in models that
include only conduction (e.g., Guo & Oh 2008; Conroy
& Ostriker 2008).
Cosmic rays from a central AGN have been invoked
to prevent catastrophic cooling of the plasma in cluster
cores, either directly via Alfven waves driven by cosmic
rays heating the plasma (e.g., Loewenstein, Zweibel, &
Begelman 1991; Guo & Oh 2008), or indirectly via convection driven by cosmic rays, the dissipation of which
heats the plasma (Chandran & Rasera 2007; Rebusco
et al. 2006). Although there is ample evidence for the
presence of cosmic rays in radio emitting bubbles in clusters (e.g., Brzan et al. 2004), it is unclear how/whether
cosmic rays can be spread throughout the cluster volume at a sufficient level for these heating mechanisms to
work. Simple hydrodynamic jets do not couple their energy to most of the ICM and instead simply drill through
it, without heating and without transporting cosmic rays
(and metals) throughout the ICM (Vernaleo & Reynolds
2006).
In this paper we show that cosmic rays, which are
likely to be centrally concentrated in clusters, can drive
efficient convection and mixing if the cosmic ray pressure is not negligible compared to the plasma pressure
(pcr /p & 0.25); this is likely the case in and around radio
bubbles. We argue that on the large scales that likely
dominate the turbulent dynamics in the ICM, cosmic
rays are effectively adiabatic rather than diffusive (i.e.,
the cosmic ray diffusion time is longer than the buoyancy time). As a result, the cosmic rays can drive a
Schwarzschild-like adiabatic convective instability. We
present a linear analysis demonstrating that, while magnetic reorientation can shut off diffusive (isobaric) cosmic ray instabilities, it cannot shut off the adiabatic
buoyancy instability driven by a negative cosmic ray entropy gradient. We then present two-fluid (plasma and
cosmic rays) numerical simulations with thermal conduction and cosmic ray diffusion along magnetic field lines.
We do not include plasma cooling in this paper, nor do
we include the heating of the thermal plasma that arises
from the excitation of short wavelength Alfven waves by
streaming cosmic rays. These are both significant omissions and preclude our results from being an accurate
representation of the plasma in cluster cores. However,
the main focus of this paper is not to solve the cooling
flow problem per se, but rather to isolate and understand the transport and turbulence properties of cluster
plasmas with realistic physics (e.g., anisotropic conduction, convection, and cosmic rays). By neglecting cooling, our calculations implicitly assume that some unspecified source of heating is preventing the rapid cooling
of the ICM. A study of cluster cores with cooling and
anisotropic conduction will be presented in a separate
paper.
The remainder of this paper is organized as follows.
In 2 we present the basic equations used in our analysis and derive the dispersion relation for linear buoyancy waves in the presence of thermal plasma and cosmic

rays. In 3 we describe our numerical simulations and

show that a steep cosmic-ray entropy gradient drives convective motions at small radii in cluster cores, and that
anisotropic thermal conduction drives convection at intermediate radii that rearranges the magnetic field structure in clusters. Readers interested in just the numerical results can skip the linear stability calculation in 2.
In 4 we summarize and discuss the implications of our
work.
2. BASIC EQUATIONS

To describe cosmic rays and thermal plasma in the

ICM, we use the two-fluid model of Drury & V
olk (1981),
modified to include anisotropic transport, gravitational
acceleration g = g r (g = d/dr, where is the gravitational potential), and a cosmic ray source term to build
up cosmic ray pressure. The equations of this model are
d
= v,
dt

(1)

( B) B
dv
=
(p + pcr ) g r,
dt
4

(2)

B
= (v B),
t

(3)

dp p d

= ( 1) Q,
dt
dt

(4)

cr pcr d
dpcr

= + (cr 1)Qc
dt
dt

(5)

and

where d/dt = /t + v is the Lagrangian time derivative,

Q = k
b(
b T )
(6)
is the heat flux, T is the plasma temperature,
= Dk
b(
b pcr )

(7)

is the diffusive flux of cosmic-ray energy (multiplied by

[cr 1]), Qc is the cosmic ray energy source term, is the
mass density, v is the common bulk-flow velocity of the
thermal plasma and cosmic rays, B is the magnetic field,

b = B/B, p and pcr are the thermal-plasma and cosmicray pressures, k is the parallel thermal conductivity, Dk
is the diffusion coefficient for cosmic-ray transport along
the magnetic field, and =5/3 and cr =4/3 are the adiabatic indices of the thermal plasma and cosmic rays,
respectively.
As mentioned in 1, we do not include radiative cooling in the energy equation (eq. [4]). In addition, we do
not include the effects of cosmic ray streaming relative to
the thermal plasma: in particular we neglect Alfven wave
heating of the thermal plasma and Alfven wave streaming in the cosmic ray energy equation (e.g., Loewenstein,
Zweibel, & Begelman 1991). This physics will be included in the future together with plasma cooling. For
the present paper we focus on the physics of convective
instabilities in clusters using a simplified but physically
reasonable model.

3
2.1. Linear Stability Analysis
We take all quantities to be the sum of an equilibrium
value plus a small-amplitude fluctuation: B = B 0 + B 1 ,
etc. We take the equilibrium velocity v 0 to vanish, set
B 0 = B0x x
+B0z z (
z = r for the cluster, and x
is chosen
such that local magnetic field lies in the x
z plane),
and take T0 , pcr0 , and 0 to be functions of z alone. We
employ a local analysis in which all fluctuating quantities
vary as eikxit with kH 1, where H is the scale
on which the equilibrium quantities vary. We consider
the limit = 8p0 /B02 1 and work in the Boussinesq
approximation, kcs (cs is the sound speed). We also
do not include the perturbed cosmic ray source term (eq.
[5]) in our linear analysis since its form is uncertain.
In terms of the plasma displacement, = iv , the perturbed magnetic field (B 1 = i[k B 0 ]) can be combined
with equation (2), to give

0 2 =

(k B 0 )2
+ ik1 + 1 g
z,
4

(8)

where 1 = p1 + pcr1 + B 0 B 1 /4 is the total-pressure

perturbation. Dotting equation (8) with k and using the
near-incompressibility condition (k /H), we find
to leading order in (kH)1 that

In the same way that we obtained equations (11)

and (12), for the cosmic rays we find that


i(2k bz z ) dpcr0
cr pcr0 M 2 z
+
pcr1 =
( + i)g
+ i
dz
cr pcr0 1
,
(13)
+
( + i)0
where = kk2 Dk is the rate at which diffusion smoothes
out variations in pcr along the magnetic field, and
d
ln(pcr0 /0cr )
dz
is the square of the Brunt-V
aisala frequency associated
with the cosmic ray pressure.
Adding equations (12) and (13), making use of equation (9), and noting that the right-hand side of equation (9) is much smaller than the individual terms on
the right-hand sides of equations (12) and (13), we find
that
"
#
2
12 (z 2k bz )
z N
1
1
1
2i kk k +
,
=
+
0
g
g
(14)
where
M 2 = (g/cr )

1 = i1 gkz /k 2
and


1 g
kkz
2
2 2
2 ,
( kk vA ) =
z
0
k

(10)

0 , and v 2 = B 2 /40 is the square of the

where kk = k b
0
A
Alfven speed.
In terms of the perturbation to the magnetic-field unit
0 B 1 /B0 ) = ikk [ b
0 (b
0
0 b
vector,
b1 ( B 1 /B0 b
)], the perturbed heat flux can be written as Q1 =
1 (b
0 T0 )+ b
0 (b
1 T0 )+ b
0 (b
0 T1 )], where we
k,0 [b
have dropped the term involving the perturbed thermal
conductivity since it is smaller than the other terms by a
factor of (kH)1 . Taking the dot product of the heat
flux with k we find that
ik Q1 = k,0 kk2 (2k bz z )

dT0
kk2 k,0 T1 ,
dz

(11)

0 and bz is the z component of b0 . Using

where k = b
equations (4) and (11), and using p1 /p0 = 1 /0 +T1 /T0 ,
we find that


ip0 (2k bz z ) d ln T0
p0 N 2 z
p1 =
+
( + i)g
+ i
dz
( + i)p0 1
,
(12)
+
( + i)0
where = ( 1)kk2 k,0 T0 /p0 is the rate at which thermal conductivity smoothes out temperature fluctuations
along the magnetic field, and
N 2 = (g/)

d
ln(p0 /0 )
dz

is the square of the Brunt-V

aisal
a frequency.

N = N 2 /( + i) + cr M 2 /( + i),

(9)

= pcr0 /p0 ,

12 = g

and

i
+ i

d ln T0
+
dz

i
+ i


d ln pcr0
,
dz

= ( + i)/( + i) + cr /( + i).
Upon substituting equation (14) into equation (10), we
obtain an equation for the plasma displacement alone,



2ikk k g
kkz
2
2 1
) z
( 2 kk2 vA
k

i
2
2
+ z N + 1 (z 2k bz ) = 0.
(15)

z
z/|k
| and
We set = 1 e
1 + 2 e2 , where e1 = k
e
2 = k
1 . After taking the dot product of equae
1 and e
2 , we obtain two equations which
tion (15) with e
can be written in matrix form as
 2
   
2
kk2 vA
0
0
1
(16)
2 = 0 .
A21
A22
Setting the determinant of the matrix on the left-hand
side of equation (16) equal to zero, we obtain the dis2
persion relation ( 2 kk2 vA
)A22 = 0, where2 A22 =

 2
2
2 kk2 vA
1 sin2 N + J12 , is the angle be,
tween k and z
2
,
J = 1 2b2z + 2bx bz kx kz /k
2

We have dropped a term on the right-hand side of expression

2
for A22 equal to 2i1 sin2 kk g 1 (bx kx kz k
bz ), which is
small for kH; in the opposite limit = kk vA + i with
2 : k g 1 : N
/ (kH)1 . Notice that kk2 vA
k

kH 1 : 1.

:: (kH)2 1 :

4
is a factor that depends on the magnetic field geome2
try and wavenumber, k
= kx2 + ky2 , and bx is the x0 .
component of b
2
While one solution to equation (16), 2 = kk2 vA
, describes waves unaffected by buoyancy, A22 = 0 corresponds to the modes modified by buoyancy,
 2

2
1 sin2 N + J12 = 0.
2 kk2 vA
(17)

In the limit that pcr /p 0, equation (17) reduces to

equation (13) of Quataert (2008). For nonzero (pcr /p),
we consider two limiting cases of equation (17). The
first is a highly diffusive limit for cosmic rays, in which
{, } {, N, 1 } kk vA . In this case, equation (17)
reduces to


d ln T0
1 dpcr0
2
= gJ
.
(18)
+
dz
p0 dz

This is a generalization of the HBI (heat-flux driven

buoyancy instability; Quataert 2008) and MTI (magnetothermal instability; Balbus 2000) in the presence
of diffusive cosmic rays (e.g., Chandran & Dennis 2006;
Dennis & Chandran 2008).
The second limit we consider is that of adiabatic cosmic
rays, with {, N, 1 } {, kk vA }. In this case,
the cosmic rays diffuse slowly compared to the buoyancy
time and behave nearly adiabatically; equation (17) then
reduces to




d
pcr0
g sin2 pcr0 d
. (19)
+
J
ln
ln
T
2 =
0
1 + cr p0 dz
0cr
dz
This implies that if the cosmic ray pressure is significant
and if cosmic ray entropy (pcr0 /0cr ) is sufficiently peaked
and decreasing outward, the plasma will become unstable to a Schwarzschild type buoyancy instability. The
entropy of cluster plasma is stably stratified according
to the Schwarzschild criterion. The thermal plasma response at small scales is nonetheless governed by the temperature gradient and not the entropy gradient. For typical cluster parameters, although the global conduction
timescale may be longer than the buoyancy timescale
(see Fig. 1), anisotropic conduction determines the local
buoyant response (i.e., & N for kr 1) via HBI/MTI
depending on the sign of temperature gradient. The adiabatic cosmic-ray instability (ACRI) described by equation (19) is different from the MTI and HBI in that
it cannot saturate by magnetic field reorientation (the
cosmic-ray driving in eq. [19] does not depend on the
field configuration term J that shows up in the thermal driving). It must lead to vigorous convection that
changes the cosmic ray entropy profile to become nearly
adiabatic.
2.2. The Cosmic Ray Diffusion Coefficient

A cosmic ray particle streaming through a magnetized

plasma is efficiently scattered in pitch angle by magnetic
fluctuations with wavelengths comparable to its Larmor
radius. An unavoidable source of magnetic fluctuations
is the self-excited streaming instability (e.g., Kulsrud &
Pearce 1969). The effect of pitch-angle scattering due to
Alfven waves is that the bulk speed of cosmic rays relative to the thermal plasma is close to the Alfven speed,

Fig. 1. Different timescales for the initial cluster model:

isothermal sound crossing time (solid line, tsnd = r/cs , where
cs = [p/]1/2 ), cooling time (dotted line; we do not include cooling
in our calculation but the cooling time is shown for comparison),
cosmic ray energy injection time (tinj = p/( 1)Qc ; short-dashed
line), HBI growth time (tHBI = [gd ln T /dr]1/2 ; long-dashed line),
conduction time (short dot-dashed line; tcond = r 2 nkB /k ), and
cosmic ray diffusion time (long dot-dashed line; tdiff = r 2 /Dk ) for
Dk = 1029 cm2 s1 .

i.e., vd vA = c2 /Lcr , where vd is the cosmic ray drift

velocity relative to the thermal plasma, is pitch-angle
scattering rate, and Lcr is cosmic ray gradient scale (Kulsrud 2005). The cosmic rays stream along the magnetic
field direction, and down the cosmic ray pressure gradient. In addition to streaming with Alfven wave packets, cosmic rays also undergo momentum-space diffusion,
which leads to spatial diffusion along the field lines with
Dk = c2 / = (vd vA )Lcr . Thus, if cosmic ray scattering is efficient and vd vA , the diffusion timescale over
scales comparable to Lcr is much longer than the Alfven
crossing time.
With the above model for cosmic ray scattering one can
self-consistently calculate the cosmic ray diffusion coefficient in terms of the plasma parameters (e.g., Loewenstein, Zweibel, & Begelman 1991). In the present calculations we do not explicitly include the effects of cosmic rays streaming with respect to the plasma. Instead,
the diffusion coefficient in equation (7) should be interpreted as an effective diffusion coefficient, taking into
account both microscopic diffusion and streaming along
turbulent magnetic field lines. The cosmic ray diffusion time due to cosmic rays streaming along random
magnetic field lines can be crudely bounded by the Alfven
crossing time (. r/vA ). Together with the fact that the
true microscopic diffusion time is much longer than the
Alfven crossing time (as argued above), this motivates
our choice of Dk = rvA for the cosmic-ray diffusion coefficient in most of our numerical simulations, where
is a factor of order unity. Somewhat arbitrarily, we take
= 0.4, but our results are insensitive to so long as

5
. 1.
This estimate of Dk (. rvA ) is consistent with the measured Galactic cosmic ray diffusion coefficient for GeV
particles, 1028 cm2 s1 (Berezinskii et al. 1990), using
typical values for magnetic field strength and cosmic ray
scale height in the Galaxy. However, at higher energies
the diffusion coefficient increases as 0.5 , where is the
cosmic ray energy (e.g., see Engelmann et al. 1990). For
a cosmic ray energy distribution function steeper than
2 (in the Milky Way it scales as 2.7 from 1 to 105
GeV), the cosmic ray pressure in equation (2) will be
dominated by the lowest energy cosmic rays, and a diffusion coefficient Dk . rvA seems appropriate for the fluid
description of cosmic rays considered here. With this
choice, the ratio of the buoyancy timescale (t2buoy r/g)
to the cosmic ray diffusion timescale (tdiff r2 /Dk ) is
tbuoy /tdiff . 1/2 . In clusters 10 . . 1000 (Govoni & Feretti 2004) so that we expect cosmic rays to be
adiabatic on relatively large length scales and thus to be
susceptible to cosmic ray driven convection when their
entropy gradient is sufficiently large. Our linear stability
analysis differs from that of Chandran & Dennis (2006)
and Dennis & Chandran (2008), in that we allow for a
background heat flux. In addition, while they obtained
the dispersion relation allowing the thermal plasma and
cosmic rays to be either both in the diffusive limit or both
in the adiabatic limit, we have argued that the more relevant case in clusters is likely that of diffusive thermal
plasma and adiabatic cosmic rays. Because of uncertainties in Dk , we have carried out numerical simulations for
different values of Dk . We find that for Dk = 1028 cm2
s1 , the cosmic rays are effectively adiabatic at all scales,
and even for Dk as high as 1029 cm2 s1 , the cosmic rays
are adiabatic on large scales, r & 10 kpc (see sec. 3.2.4).
3. NUMERICAL SIMULATIONS

We have extended the ZEUS-MP MHD code (Hayes et

al. 2006; Stone & Norman 1992a,b) to include thermal
conduction along magnetic field lines (Sharma & Hammett 2007), and have added cosmic rays as an additional
fluid diffusing along magnetic field lines. We numerically
solve equations (1)-(7). As mentioned earlier, we do not
include plasma cooling. The cosmic ray energy equation, thermal conduction in equation (4), and the cosmic
ray pressure gradient in the equation of motion are implemented in an operator split fashion with appropriate
source and transport terms. Both thermal conduction
and cosmic ray diffusion are sub-cycled.
We have tested the code extensively. The anisotropic
cosmic ray diffusion equation is analogous to anisotropic
thermal conduction. We use the method of Sharma &
Hammett (2007) which preserves positivity of pcr . Appendix shows a 1-D shock tube test, adapted from Pfrommer et al. (2006), which shows that adiabatic evolution
of cosmic rays is accurate. The thermal conductivity is
chosen to be the Spitzer value,
k =

1.84 105 5/2

T ergs1 K7/2 cm1 .
ln

(20)

Based on the discussion in 2.2 we choose Dk = 0.4rvA

for most of our calculations, where vA is the local Alfven
speed; to test the dependence on Dk , we also carry out

simulations with a constant value of Dk = 1028 , 1029

cm2 s1 .
How cosmic rays are produced and distributed in the
ICM is still poorly understood (e.g., Eilek 2003). Thus,
we use a simple phenomenological source term to drive
the cosmic ray pressure in the inner parts of the cluster.
The cosmic ray energy source term in equation (5) is
based on Guo & Oh (2008),
Qc =

M c2
4r03

r
r0

3 h

1 e(r/r0 )

(21)

We take r0 = 20 kpc and = 1.5, which leads to a centrally peaked cosmic ray entropy,
and take the cosmic
R
ray energy injection rate to be Qc 4r2 dr = 51042 erg
s1 .3 Physically, the cosmic ray energy injection rate is
presumably related
R to the accretion rate onto the central
black hole, via Qc 4r2 dr = M c2 , where is the efficiency of cosmic ray energy production. Indeed, Allen et
al. (2006) show that the mechanical luminosity of jets
can be few % of the inferred Bondi luminosity. For
M = 0.1 M yr1 , our cosmic ray energy injection rate
corresponds to 103 and thus the level of cosmicray power used here is observationally and theoretically
quite reasonable.
The cosmic-ray pressure fraction and its dependence on
radius are not that well-constrained observationally. For
a few clusters (e.g., Virgo, Perseus, and Fornax) observational constraints indicate that pcr /p . 0.2 averaged
over the cluster core (Pfrommer & Enlin 2004; Churazov et al. 2008). It is possible, however, that larger
cosmic-ray pressure fractions arise in clusters in which
AGN feedback is expected to play a particularly strong
role, such as Hydra A or Sersic 159-03 (see, e.g., Zakamska & Narayan 2003; Chandran & Rasera 2007);
cosmic-rays may also be particularly important at small
radii, and quite likely contribute significantly to the pressure in and around radio bubbles (which are associated
with a deficit of X-ray emission from the thermal plasma;
e.g., Brzan et al. 2004). In our simulations, we choose
parameters in equation (21) such that the cosmic ray
pressure is smaller than the plasma pressure even at late
times. For a larger Qc , we find that the cosmic ray pressure builds up faster than the rate at which cosmic ray
pressure can be transported outwards via convection (or
diffusion, but the latter is slow), and the cosmic ray pressure can become larger than the plasma pressure. In calculations with cooling, it is likely that a larger cosmic ray
injection rate could remain consistent with pcr /p . 0.25:
if the gas is allowed to cool then a large part of the cosmic ray energy may be channeled into plasma heating
(and thus cooling) without building up a larger cosmic
ray pressure.
To assess how effectively turbulence generated by the
HBI or ACRI mixes the plasma, we solve for the advection of a passive scalar density (e.g., a proxy for metal3 A physically more realistic model would be to include a feedback source term for cosmic rays where, instead of a fixed cosmic
ray luminosity, a fixed fraction of the instantaneous mass accretion rate is converted into cosmic ray power; this level of detail
is unnecessary for studying the basic physics of cosmic-ray driven
convection but will be included in future calculations with radiative
cooling.

6
licity) f , using
df
f

+ v f = 0.
dt
t

(22)

Appendix A shows the behavior of the passive scalar density in a shock tube test.
3.1. Simulation Parameters

The simulations are carried out in spherical (r,,) geometry with the inner boundary at rin = 1 kpc and the
outer boundary at rout = 200 kpc. Strict outflow boundary conditions are applied to the radial velocity at the
inner and outer radial boundaries. The plasma pressure
and density are held fixed at the outer boundary to prevent spurious oscillations; the plasma cooling time at rout
is longer than the Hubble time. All other plasma and cosmic ray variables are copied on ghost zones at both the
inner and outer radial boundaries. Reflective boundary
conditions are applied at the boundaries ( = 0, ).
Periodic boundary conditions are applied in the direction. A logarithmic grid is chosen in the radial direction, while the grid is uniform in and . Our fiducial run uses a 128 64 32 grid, with 0
and 0 2. With these choices, r/r = 0.042,
= 0.05, and = 0.2. We also carried out a number of 128 64 2-D (axisymmetric) simulations, and one
higher resolution, 256 128, 2-D simulation for convergence studies.
We typically initialize a weak ( > 106 everywhere)
split-monopole magnetic field with B r2 , although
in one calculation (CRM), we use a monopole field to
compare simulations with and without net magnetic flux.
In runs with cosmic rays, the initial cosmic ray pressure
is small (0.005 times the plasma pressure at rin ) and
varies as r3 ; the cosmic ray pressure builds up in time
via the source term in eq. [5]). The initial pressure is
chosen such that the plasma is in dynamical equilibrium
(d[p + pcr ]/dr = g). Initial thermal equilibrium is not
imposed.
As in Guo & Oh (2008), we use cluster parameters
relevant for Abell 2199. The gravitational potential ()
is the sum of the dark matter potential (DM ),
DM =

2GM0 ln(1 + r/rs )

,
rs
r/rs

(23)

where M0 = 3.81014M is the characteristic dark matter mass, rs = 390 kpc is the scale radius (Navarro,
Frenk, & White 1997), and the potential due to the
central cD galaxy (cD ),
p
1 + (r/rg )2 )
2 ln(r/rg +
,
(24)
cD = 4G0 rg
r/rg
where rg = 2.83 kpc, 0 = 5.63 1023 g cm3 (see
Kelson et al. 2002 for cD galaxy NGC 6166).
The ideal gas law p = nkB T is used with n = ne e =
/mp , where (e ) is the mean molecular weight per
thermal particle (electron) and n (ne ) is the total (electron) number density. We assume a fully ionized plasma
with hydrogen mass fraction X = 0.7, helium mass fraction Y = 0.28, such that = 0.62 and e = 1.18. Since
we do not include cooling, the metallicity appears only in

Fig. 2. Angle averaged plasma temperature (left) and electron

number density (right) as a function of radius at different times for
CR (upper panel) and NCR (bottom panel). Solid lines are at 1/3,
1, 3, and 9 Gyr; dotted line is the initial profile. The temperature
at 100 kpc increases at late times while density at few kpc
decreases with time.
TABLE 1
Parameters for different runs
Label
CR
NCR
CRM
CR2D
CR2D-dbl
CR28
CR29
CR30
CR30-ad

Dim.
3-D
3-D
3-D
2-D
2-D
2-D
2-D
3-D
3-D

initial B
split-M
split-M
monopole
split-M
split-M
split-M
split-M
split-M
split-M

Dk
0.4rvA
0
0.4rvA
0.4rvA
0.4rvA
1028
1029
0.4rvA
0.4rvA

CR angle
900
0
900
900
900
900
900
300
300

Half-angle around the polar axis over which CR source is applied.

fiducial run.
Although monopolar, B = 0 everywhere, including the
boundaries, since the origin is excluded from the computational
domain.

Resolution for CR2D-dbl is 256 128, double that of CR2D.

Plasma is adiabatic (k = 0) for this run.

The

the conversion of pressure into plasma temperature used

to calculate the thermal conductivity in equation (20).
To roughly match the observations (Johnstone et al.
2002), the initial temperature increases linearly from 1.5
keV at rin to 4.6 keV at rout (see Fig. 2); the electron
number density is fixed to 0.0015 cm3 and the temperature is fixed to 4.6 keV at rout at all times. The
initial density, obtained from imposing hydrostatic equilibrium, is quite a bit larger than the observed density
at rin . This is because the density obtained from hydrostatic equilibrium is extremely sensitive to the form of
the temperature profile, for which we use a simple linear
fit. However, since we do not include cooling, the large
inner density does not significantly affect our results.
3.2. Results

Table 1 summarizes the properties of our simulations.

Although we list a number of calculations in Table 1,
we focus on two 3-D simulations: CR (the fiducial run)
and NCR. Cosmic rays are not included in NCR; NCR
thus serves as a control run that allows us to isolate the
effects of cosmic rays. The aim of the rest of the simulations is to understand certain aspects of the physics
in more detail. As described earlier, for all runs except
CRM we initialize a split-monopole magnetic field. For
convergence studies, we carried out a two dimensional

7
(axisymmetric) version of CR, CR2D, and compared it
with a run with double the resolution, CR2D-dbl. All
runs, except CR28 and CR29, use Dk = 0.4rvA ; a fixed
value for Dk is chosen for CR28 and CR29 to test the
influence of Dk on our results. In runs CR30 and CR30ad, the cosmic ray source term is only applied within
300 of the pole to study angular diffusion of cosmic rays
with (CR30) and without (CR30-ad) thermal conduction
along field lines; in the rest of the calculations, cosmic
ray injection is spherically symmetric.
Our initial profiles are in dynamical equilibrium, but
not in thermal equilibrium. The plasma will remain
static if thermal conduction and cosmic rays are not included, since the magnetic field is very weak and the
plasma is stably stratified according to the Schwarzschild
criterion. However, in the presence of thermal conduction, the background temperature and density change in
time; in addition, the magnetic field lines are reoriented
by the HBI. When the cosmic ray source term is applied,
the cosmic ray entropy gradient can drive convection and
mixing due to the ACRI. In this section we discuss the
influence of these effects on the structure of the ICM.
Figure 1 shows different timescales in the initial state:
the isothermal sound crossing time (solid line) is the
shortest timescale; the plasma cooling time (dotted line)
is included for comparison with the other timescales, although cooling is not included in the simulations; the
growth-time of the HBI (long-dashed line) varies slowly
as a function of radius; the cosmic ray injection timescale
(short-dashed line), i.e., the timescale for the cosmicray source term to increase the cosmic ray pressure by
an amount comparable to the plasma pressure, increases
rapidly with radius since the cosmic ray source term is
centrally concentrated; the thermal conduction timescale
(short dot-dashed line) has a maximum at intermediate
radii and is comparable to the HBI timescale at both
rin and rout ; finally, the cosmic ray diffusion timescale
(long dot-dashed line; shown for Dk = 1029 cm2 s1 ) is
shorter than the buoyancy timescale only within 10
kpc. Since the diffusion time is 1/Dk , this implies
that cosmic rays are effectively adiabatic for smaller Dk ,
i.e., in all runs except CR29 (Table 1).
3.2.1. The Fiducial Run (CR)

We discuss the fiducial simulation (CR) in detail and

compare it with the simulation without cosmic rays
(NCR). Both simulations CR and NCR show similar
properties for radii & 30 kpc, the radius outside of which
the cosmic ray pressure is always small. Since there is
no cooling to balance heating by thermal conduction in
the initial state, the initial thermal properties will be
modified on the conduction timescale.
Figure 2 shows the angle averaged temperature and
density profiles as a function of radius for different times
for runs CR and NCR. Figure 2 shows that for run NCR,
the temperature becomes isothermal near the inner and
outer radii due to thermal conduction along initially radial magnetic field lines. Although Figure 1 shows that
the HBI timescale is always shorter than the conduction
time, the two are of the same order at both rin and rout
where thermal conduction makes the plasma isothermal
before the HBI can grow significantly. The HBI, which
is active within 30 kpc . r . 100 kpc, reorients the
magnetic field lines to be primarily perpendicular to the

Fig. 3. Time (from tH /2 to 3tH /4) and angle averaged angle between magnetic field unit vector and the radial direction
r
|i in degrees) as a function of radius: for CR (solid
(cos1 h|b
line), for NCR (dot-dashed line), for CR2D (dotted line), and for
CR2D-dbl (dashed line).

radial direction (see Fig. 3). This creates a thermal

barrier and a large temperature gradient at these radii.
The formation of such a thermal barrier is not forced by
the boundary condition since the temperature at rin is
floating; if the conduction timescale were shorter than
the HBI timescale at all radii, the plasma would become
isothermal (4.6 keV) at all radii. Run CR shows similar
behavior at larger radii but differs substantially at small
radii. Since the cosmic ray pressure provides a substantial fraction of the pressure support to balance gravity
within 30 kpc, the plasma density and pressure at small
radii in run CR is substantially smaller than in NCR.
The plasma temperature is similar in magnitude for CR
and NCR but shows a minimum at 20 kpc for CR at
late times.
Figure 4 shows the cosmic ray entropy profile (top) and
the ratio of cosmic ray pressure to plasma pressure (pcr /p,
bottom) as a function of radius for different times; we define cosmic ray entropy as pcr /cr since this is the quantity whose gradient determines the buoyant response of
adiabatic cosmic rays (see, e.g., eq. [19]). Since the cosmic ray source term is chosen to be a strong function of
radius (Qc r2.5 for r . 20 kpc), it drives convection
due to the ACRI when the cosmic ray pressure builds
up and becomes comparable to the plasma pressure. At
later times, cosmic ray injection does not continuously increase the inner cosmic ray pressure with time; instead, a
cosmic ray driven convection front spreads radially outwards. Convection drives the cosmic rays to be adiabatic (pcr /cr constant) in regions where the cosmic
ray pressure is not negligible compared to the plasma
pressure.
Figure 5 shows 2-D contour plots of turbulent velocity (absolute value of velocity) in the = plane at 9
Gyr for CR and NCR. Velocities are similar for r & 30

Fig. 4. Angle averaged cosmic ray entropy (arbitrary units;

top), and the ratio of the cosmic ray pressure to the thermal plasma
pressure (bottom) as a function of radius for 1/3, 1, 3, and 9 Gyr
(solid lines); dotted line is the initial profile. The ACRI flattens
the cosmic ray entropy profile and both pcr /cr and pcr /p increase
and move outwards in time.

CR

NCR

40

40
7

30

30
20

20
6.5
z(kpc)

6.5
10

10
0

10

10

20

5.5 20

5.5

30

30
40
0

5
20
x(kpc)

40

5
40
0

20
x(kpc)

40

Fig. 5. Contour plot (in the = plane) of Log10 v (in cm

s1 ) at 9 Gyr for CR (left) and NCR (right). Turbulent velocities
& 100 km s1 are attained in inner 20 kpc for cosmic-ray driven
convection (left). The turbulent velocities are significantly smaller
in the absence of cosmic rays. Arrows show the r projection of
the velocity unit vector.

kpc where cosmic ray pressure is negligible and fluid motions are driven by the HBI. The maximum turbulent
velocity in the HBI-dominated regions is 30 km s1 .
Run CR shows much larger turbulent velocities ( 100
km s1 ) for r . 30 kpc; note that the turbulent velocities induced by the ACRI are consistent with mixing
length theory, with Lc 4r2 vc3 , where Lc is the power
carried by convection and vc is the resulting convective
velocity. The large turbulent velocities in the presence of

the ACRI are sufficient to prevent catastrophic cooling

according to the models of Chandran & Rasera (2007).
By contrast, the turbulent velocities are extremely small
at r . 30 kpc for NCR because the plasma temperature
gradient is wiped out by conduction before the HBI can
drive any turbulence.
While the turbulent velocity vectors are roughly
isotropic at r . 30 kpc for CR, they are aligned primarily perpendicular to the radial direction at large radii
where the HBI dominates. This is consistent with the
result that while the HBI saturates by reorienting magnetic field lines perpendicular to gravity (e.g., Parrish
& Quataert 2008), the ACRI drives roughly isotropic
convection irrespective of the magnetic field geometry.
Figure 3 shows the time-averaged (from tH /2 to 3tH /4)
angle between the magnetic field vector and the radial
direction as a function of radius for runs CR and NCR;
the angle is defined with respect to the radial direction,
such that it is 00 for the initial split-monopole field. The
average angle is similar for r & 30 kpc, where the cosmic ray pressure is negligible, for both simulations. For
run CR, radii r . 30 kpc are convectively stirred by
the ACRI and the average angle between the magnetic
field and the radial direction is 550 , close to the value
expected for a uniform, random magnetic field unit vector (cos1 1/2 = 600 ). For both CR and NCR at 30 kpc
. r . 100 kpc, the magnetic field is nearly perpendicular
to the radial direction because of the HBI. The average
angle between the magnetic field unit vector and the radial direction at these radii is 750 . For r & 100 kpc
and for r . 20 kpc in run NCR, the HBI is weak since
thermal conduction wipes out the temperature gradient
before the HBI can grow significantly; as a result, the
field lines are not perpendicular to the radial direction.
Figure 6 shows angle and time averaged (from tH /2 to
3tH /4) kinetic and magnetic energy profiles as a function of radius for runs CR, CRM, and NCR. Both the
magnetic and kinetic energies are generally amplified, although to varying degrees. The kinetic energy is very
effectively amplified at small radii by the action of the
ACRI in run CR; the turbulent Mach number is 0.1
(see also Fig. 5). The magnetic energy as a function of
radius in CR is quite striking in that there is no magnetic
energy enhancement at small radii: there is a slight bump
in magnetic energy at 20 kpc but the final magnetic energy is smaller than the initial magnetic energy for r . 3
kpc. In turbulent dynamos one often finds the turbulent
magnetic energy to be of the order of turbulent kinetic
energy (e.g., Cho & Vishniac 2000); this is clearly not
the case for the run CR. Magnetic field amplification can,
however, be subtle; e.g., in local shearing box simulations
of the magnetorotational instability with no net magnetic
flux, magnetic field amplification (and associated stress
and turbulence) occurs only for Prandtl numbers exceeding unity (e.g., Fromang et al. 2007). We do not have
explicit viscosity and resistivity and it is possible that
the effective Prandtl number in our simulation is small,
so that dissipation of the initially split monopolar field
at the grid scale dominates over magnetic field enhancement by convective turbulence. To better understand
this, we have done a simulation with cosmic rays with a
net initial magnetic flux (run CRM), but with all other
properties of the simulation the same. Figure 6 shows
that in this case, the magnetic energy density is much

9
CR

x 10

40

NCR

x 10

40

30

30

z(kpc)

2
20

20

10

10

0
1

10

1
10

20

20

30

30

40
0

1.5

20
x(kpc)

40

0 40
0

0.5

20
x(kpc)

40

Fig. 7. Contour plot (at = ) of the passive scalar density, f (normalized to the initial maximum), at 9 Gyr for CR (left)
and NCR (right). The passive scalar is initialized near the origin.
Projection of magnetic field unit vector is over-plotted by arrows.
While the passive scalar diffusion is negligible for NCR, turbulent
mixing is efficient within 30 kpc for the cosmic-ray driven convection (left).
Fig. 6. Time (from tH /2 to 3tH /4) and angle averaged magnetic and kinetic energy densities (in erg cm3 ) as a function of
radius for CR (solid line), CRM (long dashed line), and NCR (dotted line). The kinetic energy density is larger than the magnetic
energy density in all cases. The initial magnetic energy density profile (short dashed line) is also shown for comparison. Run CRM,
which includes a mean field, results in a much larger amplification
of the magnetic energy as compared to CR.

larger in the inner regions as compared to CR, although

the magnetic energy is still 100 times smaller than the
kinetic energy. In this case, the magnetic field strength
in the center of the cluster is amplified to 0.1-1 G by
convective motions. The kinetic energy density profiles
are almost identical for CR and CRM, indicating that
the properties of the convection are not very different in
the two cases, although the efficiency of magnetic field
amplification differs dramatically.
At larger radii (30 kpc . r . 100 kpc), the HBI causes
amplification of the magnetic and kinetic energies in both
CR and NCR. The magnetic energy is enhanced by a factor 100 at r 60 kpc. This level of field amplification
a factor of 10, primarily of the and components
is what is required to reorient initially radial magnetic
fields into fields that are perpendicular to the radial direction (as was seen in previous HBI and MTI simulations; Parrish & Quataert 2008; Parrish & Stone 2007;
Sharma, Quataert, & Stone 2008). The kinetic energy is
also enhanced at these radii because of turbulence driven
by the HBI.
In order to study the numerical convergence of our fiducial run, we carried out an axisymmetric 2-D run analogous to CR CR2D and an axisymmetric run with
double the resolution CR2D-dbl (see Table 1); studying convergence directly with the 3D run CR would have
been computationally prohibitive, requiring 32 times
more cpu time. The results from runs CR2D and CR2Ddbl are nearly identical to each other; in particular, angle averaged plots such as those shown in Figures 2, 4,
3, and 6 are quite similar in the two cases. We have not
included these 2-D results in every figure, but to illus-

trate the basic convergence result, Figure 3 shows that

the average angle between the magnetic field unit vector
and the radial direction is quite similar for CR2D and
CR2D-dbl (which are both similar to the 3-D run CR).
It is interesting to note that the 2-D runs differ from the
3-D run CR in one important respect: the amplification
of the magnetic field at small radii is significantly larger
in 2-D than in 3-D; the magnetic field energy density
at small radii in the 2-D runs is comparable to the 3-D
run with a net magnetic flux (CRM) in Figure 6.4 As
mentioned earlier, the reason for the lack of significant
magnetic field amplification in the run CR is not entirely
clear.
3.2.2. Diffusion of a Passive Scalar

Figure 7 shows f , the passive scalar density, in the inner 40 kpc for CR and NCR at 9 Gyr, as well as the
projection of the magnetic field unit vectors. The passive scalar density is initialized to be a large number
(f = 1015 ) for r < 1.25 kpc (corresponding to four radial
zones) and is negligible (f = 1015 ) for r > 1.25 kpc.
The goal of initializing a passive scalar is to study mixing due to turbulence. Observations of clusters reveal a
metallicity distribution that is more spatially extended
than the light distribution of the central galaxy. This
may indicate turbulent transport of metals in clusters
(e.g., Rebusco et al. 2006; Rasera & Chandran 2008).
As expected, run CR with large turbulent velocities at
r . 30 kpc also results in efficient mixing. Mixing is
negligible for NCR because the inner radii (r . 30 kpc)
are isothermal and are thus not stirred by the HBI. For
a direct comparison with observations, one must include
a time dependent, spatially distributed source term in
the passive scalar equation which represents metal enrichment due to Type Ia supernovae (e.g., Rebusco et
4 By the anti-dynamo theorem, the amplification in 2D must be
transient. The dynamical time in clusters is so long, however, that
the transient can last a Hubble time!

10
100

3.7 Gyr

4.25 Gyr

0.5

0.8
0.4

50
z(kpc)

0.6
0.3
0

0.4
0.2

50

100
100

0.2

5.9 Gyr

z(kpc)

50

al. 2005); this is beyond the scope of the present paper.

Nonetheless, our results indicate that cosmic-ray driven
convection is an efficient mechanism for mixing plasma
in clusters of galaxies.
Figure 8 shows the angle averaged passive scalar density as a function of radius for run CR at 1/3, 1, 3, 9 Gyr.
Also shown is a Gaussian fit (at 9 Gyr) with a diffusion
coefficient of 1028 cm2 s1 . The passive scalar density
at 9 Gyr is flatter than the Gaussian fit at . 20 kpc,
implying that diffusion due to convection driven by cosmic rays corresponds to an effective diffusion coefficient
somewhat larger than 1028 cm2 s1 . Beyond 30 kpc,
the cosmic ray pressure is unimportant, and turbulence
is driven by the HBI. This change in the source of the
turbulence accounts for the rapid decline in the passive
scalar density at large radii in Figure 8.
To isolate just the mixing induced by turbulence driven
by the HBI, we have taken run NCR at tH /4(= 3.425
Gyr) and initialized a passive scalar density peaked at
r 53 kpc, the radius where the temperature gradient
is large and the HBI is active (see Fig. 2). The passive scalar is initialized to be 1015 at two grid points at
r = 53 kpc, = /2 and = , and negligible (1015 )
elsewhere. Figure 9 shows = snapshots of the passive
scalar density at later times. The passive scalar diffuses
more rapidly in the direction. This is because the turbulent velocities due to the HBI are larger in the direction perpendicular to gravity than they are in the radial
direction (just as the magnetic field components perpendicular to gravity are preferentially amplified). To estimate the diffusion coefficient in the r and directions, we
compare how much the passive scalar has spread in the
two directions; the full width at half maximum (FWHM)

10.8 Gyr

0.025

0.08

0.02

0.06

0.015

0.04

0.01

0.02

0.005

50

Fig. 8. Angle averaged passive scalar density (Log10 f ; normalized to the initial maximum) as a function of radius for CR at 1/3,
1, 3, 9 Gyr (solid lines); the initial profile, shown by the dotted
line, very close to the y axis. Passive scalar density decreases as
it spreads out with time. For comparison, the dashed line shows
a Gaussian fit with a diffusion coefficient D = 1028 cm2 s1 at 9
Gyr ( exp[r 2 /6Dt]).

0.1

100
0

50
x(kpc)

100

50
x(kpc)

100

Fig. 9. Contour plot (at = ) of the passive scalar density

(f , normalized to the initial maximum) at 3.7, 4.25, 5.9, 10.8 Gyr
for the run NCR. The passive scalar density is initialized in a small
region (two grid points along r, , ) centered at 53 kpc at 3.425
Gyr, and is negligible everywhere else. Turbulent mixing due to the
HBI is faster in the direction compared to the radial direction.

for f along r and at 9 Gyr is 10, 40 kpc, respectively.

For comparison, the FWHM at 9 Gyr for f for run CR
shown in Fig. 8 is 60 kpc. Thus the diffusion coefficient due to the HBI alone is 2 (perpendicular to
gravity) and 50 (parallel to gravity) times smaller than
the diffusion coefficient due to the ACRI in run CR. Although these precise numerical values likely depend on
the detailed parameters of our simulations, the fact that
the HBI primarily induces turbulence and mixing in the
plane perpendicular to gravity is a generic result.
3.2.3. Heat Flux modified by the HBI and ACRI

Many 1-D models of clusters parameterize thermal conduction by its ratio to the Spitzer value (Zakamska &
Narayan 2003; Chandran & Rasera 2007; Guo & Oh
2008). However, our simulations show that, because of
plasma instabilities that operate in clusters (e.g., HBI
and ACRI), a reduction of the conductivity by a fixed
factor is not applicable (see also Parrish & Stone 2007;
Parrish & Quataert 2008; Sharma, Quataert, & Stone
2008). At large radii in cluster cores, where the cosmic ray pressure is negligible, the HBI can orient field
lines perpendicular to the radial direction, but at small
radii where cosmic rays can be significant, the magnetic
field may be significantly more radial. For example, for
run CR the average angle of the magnetic field relative
to the radial direction for r . 30 kpc is 550 (see
Fig. 3), corresponding (roughly) to a reduction factor
of (
b r)2 1/3. For 30 kpc . r . 100 kpc, however,
the turbulence is dominated by the HBI, and the average
angle between the magnetic field and the radial direction

11

Fig. 10. Angle averaged cosmic ray entropy (upper panel; arbitrary normalization) and the ratio of cosmic ray to plasma pressure
(lower panel) as a function of radius for the runs CR28 (left) and
CR29 (right), which use a fixed cosmic ray diffusion coefficients of
Dk = 1028 and 1029 cm2 s1 , respectively. The solid lines show
profiles at 1/3, 1, 3, 9 Gyr, with the entropy and pressure ratio
increasing in time; the initial profiles are shown with dotted lines.
Profiles for CR28 look similar to profiles for CR in Figure 4. The
profiles for CR29 are different in that pcr /cr and pcr /p increase
towards a maximum at the intermediate radii; this is because the
cosmic-rays are no longer adiabatic for large Dk .

is 750 , corresponding to a reduction factor of 0.07.

An even more subtle result is that the HBI can change
the background temperature gradient by forming thermal barriers (see Fig. 2), which will be absent with
isotropic conduction. For example, the temperature gradient at 30 kpc . r . 100 kpc for CR and NCR at late
times is 3 times larger than the initial temperature
gradient. Thus, the HBI not only reduces the conductive
heating by a factor of (
b r)2 0.07, it also increases it
by making the temperature gradient larger by a factor
3. Approximating the conductivity of a magnetized
plasma by a constant factor with respect to the Spitzer
value misses all of this interesting dynamics. Whether
this is important in real clusters with radiative cooling
and various sources of heating remains to be seen.
3.2.4. Runs with larger Dk (CR28 & CR29)

We have also carried out 2-D simulations with larger

cosmic ray diffusion coefficients (Dk ), since the value of
the cosmic ray diffusion coefficient is poorly constrained
(see 2.2). Run CR28 uses Dk = 1028 cm2 s1 , the
cosmic ray diffusion coefficient estimated for GeV cosmic
rays in the Galaxy. Run CR29 uses Dk = 1029 cm2
s1 . All other parameters and initial conditions are same
as the fiducial run. Since we are comparing these 2-D
simulations with CR, which is a 3-D simulation, we have
verified that the run CR2D gives results similar to the
3-D results presented here; in particular, the profiles for
pcr /cr and pcr /p are identical to the profiles for CR
shown in Figure 4.
Figure 10 shows the angle averaged cosmic ray entropy
(pcr /cr ) and the ratio of the cosmic ray pressure to
the plasma pressure (pcr /p) as a function of radius for
CR28 and CR29. The profiles for CR28 and CR (see
Fig. 4) look similar; cosmic ray entropy and pressure
are slightly more radially spread out for CR28. The
profiles for CR29 are quite different. The entropy and
pressure ratio profiles have a peak at 20 kpc; this is
the radius beyond which the cosmic ray diffusion time is
longer than the buoyancy timescale (see Fig. 1). The
ratio pcr /p is smaller in CR29 as cosmic rays are spread
out over a larger volume by diffusion. The cosmic ray

entropy increases outwards for r . 20 kpc since cosmic

ray diffusion, and not convection driven by cosmic rays,
dominates the outward cosmic ray transport. This is different from CR and CR28 where cosmic ray diffusion is
sub-dominant. Even in CR29, cosmic rays are effectively
adiabatic for r & 20 kpc; cosmic-ray driven convection
is absent at these radii, however, because the cosmic ray
pressure is not large enough to drive the ACRI (see Fig.
10). At smaller radii, the cosmic rays are nearly isobaric
because of rapid diffusion, and the system is formally
unstable to the CR mediated version of the MTI (eq.
[18]). However, because the field lines are nearly radial
at these radii, the growth-rate of the CRMTI is quite
slow and we do not see any indications that it develops
in our simulations.
To explicitly study the possibility of the ACRI setting
in at larger radii in the cluster core, we carried out a
simulation with the cosmic ray source term (eq. [21])
three times larger than in run CR29. This larger source
term increases the CR pressure at large radii, and at late
times pcr /p is large enough to drive the ACRI. The turbulent velocities are 100 km s1 at r 20 30 kpc,
where the cosmic rays are effectively adiabatic in spite
of the large Dk . Thus, even in the presence of rapid
cosmic ray diffusion, the ACRI can set in at large radii
where the cosmic rays are adiabatic, provided the cosmic
ray pressure is sufficiently large; this may naturally occur in the vicinity of cosmic-ray filled buoyant bubbles.
More generally, our results demonstrate that so long as
Dk . 1 3 1028 , cm2 s1 , cosmic rays will behave
effectively adiabatically throughout the cluster core and
bulk transport by convection and other mechanisms will
dominate the diffusive transport.
3.2.5. Runs with Cosmic ray Sources at the Poles (CR30 &
CR30-ad)

It is very unlikely that cosmic rays in clusters are injected spherically symmetrically. Instead, the injection
likely occurs preferentially in the polar direction. To
study the resulting physics in this case, we carried out
3-D simulations in which the cosmic ray source term is
applied only within 300 of the pole: CR30 and CR30-ad.
Except for this difference all parameters for run CR30 are
the same as run CR. Run CR30-ad differs from CR30 in
that the plasma is adiabatic, i.e., thermal conduction is
not included. One of the aims of these simulations is to
show the dramatic differences that result from including
anisotropic thermal conduction (relative to a more typical adiabatic simulation). Cluster plasmas are observed
to be stable to adiabatic convection because the entropy
increases outwards (e.g., Piffaretti et al. 2005). However, convection in an anisotropically conducting plasma
depends on the temperature gradient, and not the entropy gradient, and the system is unstable independent
of the sign of the temperature gradient. This makes it
much easier to mix a thermally conducting plasma than
an adiabatic plasma. In clusters, this implies that turbulence produced by external means, e.g., the ACRI, wakes
due to galaxy clusters, etc., may be an effective way of
mixing the thermal plasma.
Figure 11 shows contour plots ( = snapshot) of the
ratio of cosmic ray pressure to plasma pressure (pcr /p) at
1/3, 1, 3, 6 Gyr, for CR30 and CR30-ad. For CR30, the
cosmic rays become unstable to the ACRI in the polar

12
1/3 Gyr

3 Gyr

1 Gyr

6 Gyr

40
0.5

CR30
z(kpc)

20
1
0
1.5

20

40
40
CR30ad

2.5

z(kpc)

20
3
0
3.5
20
40
0

4
20
x(kpc)

40 0

20
x(kpc)

40 0

20
x(kpc)

40 0

20
x(kpc)

40

Fig. 11. Contour plots within 40 kpc, in the = plane,

of Log10 (pcr /p) for runs CR30 (top) and CR30-ad (bottom) at
t = 1/3, 1, 3, and 6 Gyr (from left to right). The ratio
pcr /p is not shown if it is smaller than 104 . The adiabatic
plasma in CR30-ad (bottom) artificially suppresses the angular
and radial mixing of the relativistic and thermal plasma that
is present in the simulations with anisotropic thermal conduction (CR30; top). For movies corresponding to this figure, see:
http://astro.berkeley.edu/psharma/clustermovie.html.

region. The resulting turbulence is able to convectively

mix the plasma, not only in the unstable radial direction,
but also in the marginally stable direction. Instead of
cosmic rays being confined only to the = 300 cone, convection effectively mixes plasma in both the radial and
angular directions. In addition, at radii beyond 20 kpc
where the temperature gradient is appreciable (see Fig.
2), the HBI mixes material primarily in the direction,
as seen by the oriented fingers at late times in Figure
11. For CR30, pcr /p 0.02 at 3 Gyr in the equatorial
region for r . 10 20 kpc. Although cosmic rays are
still dominant near the pole, convection brings a nonnegligible amount cosmic rays into the equatorial region.
In comparison, there is no sign of convective overshoot
in CR30-ad because convective motions in the thermal
plasma near the equator are strongly stabilized by a large
positive plasma entropy gradient. The polar cosmic ray
dominated plasma does expand somewhat in both r and
as it becomes over-pressured. However, the value of
pcr /p at 3 Gyr is . 104 everywhere in the equatorial
plane for CR30-ad.
4. SUMMARY & ASTROPHYSICAL IMPLICATIONS

The X-ray emitting plasma in clusters of galaxies is

hot (T 1 10 keV) and dilute (n 0.001 0.5 cm3 ),
so that the transport of heat and momentum along magnetic field lines can be energetically and dynamically important. In addition, jets launched by a central AGN
produce relativistic plasma (cosmic rays), which are observed in part as bubbles of radio emission associated
with deficits of thermal X-ray emission (X-ray cavities;
e.g., Brzan et al. 2004). In this paper we have studied
the transport properties of an ICM composed of cosmic
rays and thermal plasma. We have argued that cosmic
ray diffusion is likely to be slow because of scattering by
self-generated Alfven waves (2.2); as a result, the cos-

mic rays are adiabatic on moderately large length scales

& 1-10 kpc. More concretely, cosmic rays are adiabatic
on the scale of cluster cores, so long as their parallel diffusion coefficient satisfies Dk . 1029 cm2 s1 .5
It is now well established that anisotropic conduction
and anisotropic cosmic ray diffusion can dramatically
modify buoyancy instabilities in low collisionality systems, producing qualitatively new instabilities such as
the MTI, HBI, and their cosmic ray counterparts (e.g.,
Balbus 2000; Chandran & Dennis 2006; Parrish & Stone
2007; Quataert 2008; Parrish & Quataert 2008). Nonlinear studies of these instabilities (including those in
this paper) have demonstrated that they saturate by approaching a state of marginal stability to linear perturbations, just as in hydrodynamic convection. However,
in a magnetized plasma, there is an additional degree of
freedom that is not present in hydrodynamic convection,
namely the local direction of the magnetic field. The
primary mechanism by which these diffusive buoyancy
instabilities saturate is by rearranging the magnetic field
lines, so that the linear growth rate becomes extremely
small (see Fig. 3). This is different from entropy gradient driven convection in adiabatic fluids, which saturates
by producing convection that wipes out strong entropy
gradients. Even nonlinearly, most of the energy flux in
systems unstable to the MTI and HBI is transported by
thermal conduction, rather than convection. Moreover,
the saturation of these instabilities is quasilinear in the
sense that the saturated magnetic energy is proportional
to the initial magnetic field energy (Sharma, Quataert,
& Stone 2008).
In this paper we have shown analytically and through
numerical simulations that when cosmic rays have appreciable pressure, pcr /p & 0.25, and an outwardly decreasing entropy (pcr /cr ), they can drive strong convection
and mixing in clusters of galaxies. This adiabatic cosmic
ray instability (ACRI) in the central regions of clusters of
galaxies is a cosmic ray analogue of hydrodynamic convection familiar in the context of stars and planets. In
particular, the nonlinear saturation of adiabatic cosmic
ray convection is similar to that of hydrodynamic convection, and thus quite different from the saturation of
the MTI and HBI. Our simulations of cluster cores also
provide insight into the global saturation of the HBI at
radii in clusters where the cosmic ray pressure is negligible ( 30 100 kpc in our models), and thus where
the only convective instability is that driven by the background conductive heat flux. More specifically, the primary results of this paper include:
If the cosmic ray entropy decreases outwards
and if pcr /p & 0.25, convection driven by the
ACRI sets in. In the saturated state, the cosmic
ray entropy profile becomes nearly constant
in the region with significant cosmic ray pressure (Fig. 4). The resulting turbulent velocities
are consistent with mixing length theory, with vc
5 A purely thermal electron-ion plasma can also show an adiabatic, rather than diffusive, response even in the presence of
rapid electron thermal conduction; this occurs if electrons and
protons are not collisionally coupled on the buoyancy timescale.
We find, however, that even in the outer parts of clusters, the
electron-proton energy exchange time is shorter than the buoyancy
timescale and thus the MTI/HBI limits are appropriate.

13
100(Lc /1043 erg s1 )1/3 (n/0.1 cm3)1/3 (r0 /20kpc)2/3First, to generate the ACRI, we have injected cosmic
rays using a subsonic source term at small radii. In rekm s1 where Lc is the total power supplied to
ality a significant fraction of the cosmic rays produced
cosmic rays and r0 is the pressure height scale
by an AGN are expected to be produced in a supersonic
of cosmic rays. The ACRI generates turbulent
jet shocking against the ICM. The spatial distribution
motions more effectively in cluster cores than the
of cosmic rays produced by jets is poorly understood.
HBI alone.
The intuition drawn from our simulations should apply
The ACRI drives roughly isotropic convection with
as long the source of cosmic rays ultimately produces a
the average angle between the field lines and the
centrally concentrated bubble of relativistic plasma that
radial direction 550 ; by contrast, the HBI generexpands subsonically.
ates magnetic field lines that are primarily in the
Our calculations intentionally do not include plasma
and directions (Fig. 3), shutting off the radial
cooling. Instead of trying to solve the cooling flow probconduction of heat. The effective radial conductivlem, our goal has been to study the basic physics of buoyity of a cluster plasma thus depends sensitively on
ancy instabilities in the combined relativistic + thermal
which of these instabilities operates at a given loplasma, implicitly assuming that some heating process
cation, and may not be adequately approximated
is preventing catastrophic cooling of the plasma. Our
as a fixed fraction of the Spitzer value throughout
calculations also do not include anisotropic ion viscosthe cluster.
ity which is 40 times smaller than electron thermal
conductivity. Finally, we do not treat the effects of cos We have quantified the mixing of a passive scalar
mic ray streaming with respect to the thermal plasma
by the ACRI and HBI: the ACRI produces roughly
from first principles, although our choice of the cosmic
isotropic mixing with a turbulent diffusion coeffiray diffusion coefficient qualitatively accounts for limits
cient D & 1028 cm2 s1 (Fig. 7); mixing length
on cosmic ray streaming produced by self-excited Alfven
1/3
theory predicts that D vc Lc . At larger
waves (see 2.2). In future work, we intend to include all
radii, only the HBI operates and the mixing is priof the above effects, which will provide a more quantitative model of plasma in cluster cores.
marily in the and directions, rather than in the
radial direction (Fig. 9). Both the ACRI and the
Finally, we note that in a full cosmological context,
HBI may contribute to mixing metals in clusters
galaxy clusters will contain a large number of galaxies
and other dark matter substructure. The motion of such
by redistributing, in both radius and angle, metals
produced by Type 1a supernovae. Some observabound objects through the ICM will reorient the magtions of metallicity gradients in clusters have innetic field and generate downstream turbulence. The
interplay between this turbulence and that generated by
ferred mixing at levels comparable to those found
here (e.g., Rebusco et al. 2006).
the instabilities studied in this paper is worth investigating in detail in future work. This interaction may create
It is considerably easier to mix thermal plasma
a magnetic dynamo in the ICM that is more effective
in the presence of anisotropic thermal conduction,
than that produced by the HBI alone: galaxies moving
since the plasma is formally always buoyantly unthrough the ICM will comb out the magnetic field lines
stable and thus already prone to mixing! By conin the radial direction, while the HBI will amplify the
trast, treating the plasma as adiabatic (i.e., ignorfield and generate a strong perpendicular magnetic field
ing thermal conduction) results in an artificially
component from the seed radial field created by galactic
stabilizing entropy gradient in cluster plasmas.6 As
wakes.
a concrete example of these effects, we have demonstrated that cosmic rays initially injected into the
polar regions can be partially mixed to the equator
Support for this work was provided by NASA through
by convective overshooting in the ACRI and HBI
Chandra Postdoctoral Fellowship grant numbers PF8unstable regions (Fig. 11); this effect is largely ab90054 and PF7-80049 awarded by the Chandra X-ray
sent in simulations that treat the plasma as adiaCenter, which is operated by the Smithsonian Astrophysbatic. If wave heating due to cosmic ray streaming
ical Observatory for NASA under contract NAS8-03060.
or heating due to Coulomb interactions is imporE. Q. was supported in part by the David and Lucile
tant in clusters (e.g., Guo & Oh 2008), a mechaPackard Foundation, NSF-DOE Grant PHY-0812811,
nism similar to that described here may be crucial
and NSF ATM-0752503. B. C. was supported in part
in redistributing cosmic rays throughout the clusby NASA grant No. NNG 05GH39G and NSF grant No.
ter volume. More generally, to study the mixing
AST 05-49577. We thank the Laboratory for Computaproduced by external sources of turbulence such as
tional Astrophysics, University of California, San Diego,
galactic wakes or cosmic-ray filled bubbles, we susfor developing ZEUS-MP and providing it to the commupect that anisotropic thermal conduction must be
nity. This research was supported in part by the National
accounted for, so that the buoyant response of the
Science Foundation through TeraGrid resources provided
thermal plasma is correctly represented.
by NCSA and Purdue University. The simulations reported in the paper were carried out on the Abe cluster
Having summarized our primary results, we now deat NCSA and the Steele cluster at Purdue University.
scribe several caveats and directions for future research.

14
6 Even including isotropic thermal conduction reduces the stabilizing effect of the entropy gradient; it does not, however, capture
the MTI/HBI, which are driven by anisotropic thermal conduction

along magnetic field lines.

REFERENCES
Allen, S. W., Dunn, R. J. H., Fabian, A. C., Taylor, G. B., &
Reynolds, C. S. 2006, MNRAS, 372, 21
Balbus, S. A. 2000, ApJ, 534, 420
Berezinskii, V. S., Bulanov, S. V., Dogiel, V. A., & Ptuskin, V.
S., Astrophysics of Cosmic Rays, Amsterdam:North-Holland,
1990, Ed. Ginzburg, V. L.
Bertschinger, E. & Meiksin, A. 1986, ApJ, 306, L1
Binney, J. & Tabor, G. 1995, MNRAS, 276, 663
Brzan, L., Rafferty, D. A., McNamara, B. R., Wise, M. W., &
Nulsen, P. E. J. 2004, ApJ, 607, 800
Braginskii, S. I., Reviews of Plasma Physics, Vol. 1, Consultants
Bureau, 1965
Chandran, B. D. G. & Dennis, T. J. 2006, ApJ, 642, 140
Chandran, B. D. G. & Rasera, Y. 2007, ApJ, 671, 1413
Cho, J. & Vishniac, E. T. 2000, ApJ, 538, 217
Churazov, E., Forman, W., Vikhlinin, A., Tremaine, S., Gerhard,
O., & Jones, C. 2008, MNRAS, 388, 1062
Ciotti, L. & Ostriker, J. P. 2001, ApJ, 2001, 551, 131
Conroy, C. & Ostriker, J. P. 2008, ApJ, 681, 151
Dennis, T. & Chandran, B. D. G. 2009, ApJ, 690, 566
Drury, L. & V
olk, H. 1981, ApJ, 248, 344
Eilek, J. A. 2003, Proc. of The Riddle of Cooling Flows in
Galaxies and Clusters of Galaxies, E13, Ed. T. H. Reipeich, J.
C. Kempner, & N. Soker
Engelmann, J. J. et al. 1990, A&A, 233, 96
Fabian, A. C. 1994, ARA&A, 32, 277
Fromang, S., Papaloizou, J., Lesur, G., & Heinemann, T. 2007,
A&A, 476, 1123
Govoni, F. & Luigina, F. 2004, Int. J. Mod. Phys. D, 13, 1549
Guo, F. & Oh, S. P. 2008, MNRAS, 384, 251
Hayes, J. C. et al. 2006, ApJS, 165
Johnstone, R. M., Allen, S. W., Fabian, A. C., & Sanders, J. S.
2002, MNRAS, 336, 299
Kelson, D. D. et al. 2002, ApJ, 576, 720
Kulsrud, R. M. & Pearce, W. P. 1969, ApJ, 156, 445

Kulsrud, R. M., Plasma Physics for Astrophysics 2005, Princeton

Univ. Press
Loewenstein, M., Zweibel, E. G., & Begelman, M. C. 1991, ApJ,
377, 392
Navarro, J. F., Frenk, C. S., & White, S. D. M. 1997, ApJ, 490,
493
Parrish, I. J. & Stone, J. M. 2007, ApJ, 664, 135
Parrish, I. J. & Quataert, E. 2008, ApJ, 677, L9
Parrish, I. J., Stone, J. M., & Lemaster, M. N. 2008, ApJ, 688,
905
Peterson, J. R. et al. 2003, ApJ, 590, 207
Pfrommer, C. & Enlin, T. A. 2004, A&A, 413, 17
Pfrommer, C. et al. 2006, MNRAS, 367, 113
Piffaretti, R. et al. 2005, A&A, 433, 101
Quataert, E. 2008, ApJ, 673, 758
Rasera, Y. & Chandran, B. 2008, ApJ, 685, 105
Rasera, Y., Srivastava, K., & Chandran, B. 2008, ApJ, 689, 825
Rebusco, P. Churazov, E., B
ohringer, H., & Forman, W. 2005,
MNRAS, 359, 1041
Rebusco, P., Churazov, E., B
ohringer, H., & Forman, W. 2006,
MNRAS, 372, 1840
Reynolds, C. S., McKernan, B., Fabian, A. C., Stone, J. M., &
Vernaleo, J. C. 2005, MNRAS, 357, 242
Rosner, R. & Tucker, W. H. 1989, ApJ, 338, 761
Schwarzschild, M., Structure and Evolution of Stars, Princeton
University Press, 1958
Sharma, P. & Hammett, G. W. 2007, J. Comp. Phys., 227, 123
Sharma, P., Quataert, E., & Stone, J. M. 2008, MNRAS, 389,
1815
Stone, J. M. & Norman, M. L. 1992, ApJS, 80, 753
Stone, J. M. & Norman, M. L. 1992, ApJS, 80, 791
Vernaleo, J. C. & Reynolds, C. S. 2006, ApJ, 645, 83
Zakamska, N. & Narayan, R. 2003, ApJ, 582, 162

APPENDIX
NUMERICAL TESTS

Cosmic ray shock tube

We have tested the adiabatic implementation of cosmic rays with a 1-D shock tube problem discussed in Pfrommer
et al. (2006) and in Rasera & Chandran (2008). Like Rasera & Chandran (2008), we also use 1024 grid points.
Thermal conduction and cosmic ray diffusion are absent for this test problem. The left (1 < x < 1.5) and right
(1.5 x < 2) states are given by (L , vL , pL , pcrL ) = (1, 0, 6.7 104 , 1.3 105 ) and (R , vR , pR , pcrR ) = (0.2, 0,
2.4 102 , 2.4 102), respectively. Figure 12 shows profiles at t = 0 (solid line) and at t = 4.4 104 (shorter than the
crossing time; points). The profiles match very well with the analytic result and with Fig. 3 of Rasera & Chandran
(2008). In addition we also test advection of a passive scalar density governed by equation (22). The passive scalar
density shows a discontinuity at the location of the contact discontinuity; volume integrated f is not conserved but
volume integrated f is.

15

Fig. 12. Profiles for different fluid variables for the shock tube test: initial profiles (solid line) and profiles at t = 4.4 104 (points).
While the shock is resolved by 4 points, contact discontinuity requires more grid points to be resolved (this is a feature of all methods that
do not solve the full Riemann problem).