a

r
X
i
v
:
1
2
0
9
.
4
0
6
3
v
2


[
a
s
t
r
o
-
p
h
.
H
E
]


9

S
e
p

2
0
1
3
Draft version September 10, 2013
Preprint typeset using L
A
T
E
X style emulateapj v. 5/2/11
ELECTRON-POSITRON FLOWS AROUND MAGNETARS
Andrei M. Beloborodov
Physics Department and Columbia Astrophysics Laboratory, Columbia University, 538 West 120th Street New York, NY 10027;
[email protected]
Draft version September 10, 2013
ABSTRACT
The twisted magnetospheres of magnetars must sustain a persistent ow of electron-positron plasma.
The ow dynamics is controlled by the radiation eld around the hot neutron star. The problem of
plasma motion in the self-consistent radiation eld is solved using the method of virtual beams.
The plasma and radiation exchange momentum via resonant scattering and self-organize into the
radiatively locked outow with a well-dened, decreasing Lorentz factor. There is an extended zone
around the magnetar where the plasma ow is ultra-relativistic; its Lorentz factor is self-regulated
so that it can marginally scatter thermal photons. The ow becomes slow and opaque in an outer
equatorial zone, where the decelerated plasma accumulates and annihilates; this region serves as a
reector for the thermal photons emitted by the neutron star. The e

ow carries electric current,

which is sustained by a moderate induced electric eld. The electric eld maintains a separation
between the electron and positron velocities, against the will of the radiation eld. The two-stream
instability is then inevitable, and the induced turbulence can generate low-frequency emission. In
particular, radio emission may escape around the magnetic dipole axis of the star. Most of the ow
energy is converted to hard X-ray emission, which is examined in the accompanying paper.
Subject headings: plasmas stars: magnetic elds, neutron X-rays
1. INTRODUCTION
The observed activity of magnetars is believed to be
caused by their surface motions, which are driven by
strong internal stresses. The magnetosphere is anchored
in the neutron star and twisted by the surface motions,
resembling the behavior of the solar corona. As a result
it becomes non-potential, B = 0, and threaded by
electric currents (Thompson et al. 2002; Beloborodov
& Thompson 2007). The currents ow along B, i.e.,
the twisted magnetosphere remains nearly force-free,
j B = 0. Numerical models of dynamic twisted magne-
tospheres of magnetars (Parfrey et al. 2012, 2013) show
how the twist creates spindown anomalies and initiates
giant ares when the magnetosphere is overtwisted and
loses equilibrium.
Free energy stored in the magnetospheric twist is grad-
ually dissipated through the continual electron-positron
discharge that sustains the electric current. As a result,
the magnetosphere tends to untwist on the character-
istic ohmic timescale of a few years. Electrodynamics of
untwisting is quite peculiar: whenever the crustal mo-
tions stop or slow down, the electric currents tend to
be quickly removed from the magnetospheric eld lines
with small apex radii R
max
(Beloborodov 2009). Cur-
rents have longest lifetimes on eld lines with R
max
R
(where R 10 km is the star radius) and form an ex-
tended j-bundle (Figure 1). The j-bundle has a sharp
boundary, which gradually shrinks toward the magnetic
dipole axis. In particular, the footprint of the j-bundle
on the neutron star surface, which may be observed as
a hot spot, shrinks with time. Shrinking hot spots were
indeed reported in transient magnetars whose magne-
tospheres were temporarily activated and then gradually
relaxed back to the quiescent state (see Figure 5 in Be-
loborodov 2011a and refs. therein).
The j-bundle must be lled with relativistic plasma
that carries the electric current j = (4/c) B. The
plasma is continually created by e

discharge near the

star and must expand along the extended eld lines. The
plasma emits persistent nonthermal emission, convert-
ing the dissipated twist energy to radiation. In the ac-
companying paper (Beloborodov 2013), we calculate the
spectrum of produced radiation and show that it forms
the observed hard X-ray component with a peak around
1 MeV. The present paper studies in more detail the dy-
namics of the e

plasma.
Previously, semi-transparent plasma ows around
magnetars were invoked to explain the deviation of the
observed 1-10 keV emission from a thermal spectrum
(Thompson et al. 2002). It was usually assumed that
the magnetar corona is lled with positive and negative
charges that are counter-streaming with mildly relativis-
tic speeds. The counter-streaming picture was motivated
by the fact that electric current j must ow along the
twisted magnetic eld lines. The coronal plasma must
be nearly neutral; it can easily carry the required cur-
rent if the opposite charges with densities n
+
= n

ow
in the opposite directions, so that j = e(v
+
n
+
v

)
where v
+
v

< 0. The velocities v

are free parameters in

this phenomenological model, which may be adjusted so
that resonant scattering of thermal X-rays in the corona
reproduces the 1-10 keV part of the magnetar spectrum
(e.g. Fern andez & Thompson 2007; Nobili et al. 2008;
Rea et al. 2008).
The counter-streaming model has, however, a prob-
lem. Note that the thermal radiation of the magne-
tar (h 1 keV) is resonantly scattered at large radii
r 10R 100 km where B 10
11
10
12
G.
1
In this
region, the plasma is strongly pushed by radiation away
from the star and the counter-streaming model needs an
1
This assumes that photons are scattered by electrons, not ions.
2
Fig. 1. Snapshot of a slowly untwisting magnetosphere. In this example, a global twist with a uniform amplitude = 0.2 was
implanted into the dipole magnetosphere at t = 0, and the snapshot shows the magnetosphere at t 1 yr. Details of the calculations
are described in Beloborodov (2009). The plane of the gure is the poloidal cross section of the magnetosphere. The black curves are the
poloidal magnetic eld lines. The magnetosphere is symmetric about the vertical axis and the equatorial plane; therefore, the gure only
shows one quarter of the poloidal cross section. The neutron star is shown by the black circle (radius R 10 km). Left panel: Current
density j normalized to BR/c. The region from which currents have been pulled into the star (the potential cavity with j = 0) is shown
in white. The boundary between the cavity and the j-bundle (magenta curve) expands with time, i.e. the j-bundle shrinks toward the
vertical axis. Right panel: Twist amplitude at the same time. The twist amplitude is dened for each closed eld line as the azimuthal
displacement of its footpoint in the southern hemisphere relative to the footpoint in the northern hemisphere.
electric eld E

that forces charges of the right sign to

move toward the star against the radiative drag. At the
same time, the electric eld acts on the opposite charges
and accelerates them away from the star, cooperating
with the radiative push. In the presence of e

plasma
(which is inevitably created near magnetars), the out-
ward acceleration generates relativistic particles, and no
self-consistent solution exists for the mildly-relativistic
counter-streaming model.
In this paper (and the accompanying paper Be-
loborodov (2013)), we develop a dierent picture of
plasma circulation in the magnetar corona. It is schemat-
ically shown in Figure 2. The outer corona is inevitably
lled with e

pair plasma of a high density n, which is

larger than j/ec by the multiplicity factor M1; in
this respect it resembles the ow along the open eld lines
of a rapidly rotating, strongly magnetized neutron star
(e.g. Hibschman & Arons 2001; Thompson 2008a; Medin
& Lai 2010). Pairs are created in the adiabatic zone
B > 10
13
G where the ow energy is reprocessed into par-
ticles with Lorentz factors 20 (Beloborodov 2013);
their multiplicity M 10
2
is basically set by energy con-
servation and mainly controlled by the discharge voltage.
Both electrons and positrons outow from the magnetar,
and radiation pressure forces the particles to accumulate
in the equatorial plane of the magnetic dipole, where
they annihilate. The required current j = (c/4) B
is sustained in the outow by a moderate electric eld
E

. This eld is self-consistently generated to main-

tain a small dierence between the velocities of the
charges, (v
+
v

)/v

M
1
1, so that the condi-
tion e(n
+
v
+
n

) = j is satised with n
+
n

and
v
+
v

> 0. In the simplest, two-uid model (Section 3)

the velocities v

tend to be locked by the balance of

two forces, electric and radiative.
The coronal outow signicantly changes the radiation
it interacts with via scattering. The problem of out-
ow dynamics can be formulated as a problem of self-
consistent radiative transfer where particles and photons
exchange energy and momentum as they ow away from
the neutron star. This problem is solved in this paper
using a specially designed numerical method.
The paper is organized as follows. Section 2 discusses
the creation of e

pairs and their circulation in the inner

and outer magnetosphere. Section 3 presents the model
of a radiatively locked outow in its simplest version us-
ing a two-uid description and assuming an optically thin
magnetosphere. Section 4 discusses the two-stream in-
stability in the e

ow and the origin of low-frequency

emission from magnetars. Then, in Sections 5 and 6
we formulate and solve the full problem where an out-
ow with a broad momentum distribution function and
signicant optical depth interacts with the neutron-star
radiation. The numerical method and results are de-
scribed in Section 6. Our conclusions are summarized in
Section 7.
2. CREATION AND CIRCULATION OF E

PAIRS
The creation of e

pairs by an accelerated particle is

a two-step process: the particle generates a high-energy
photon (via resonance scattering) and then the photon
converts to an e

pair in the strong magnetic eld. An

accelerated electron (or positron) can resonantly scatter
photons of energy once it reaches the Lorentz factor
required by the resonance condition (1 cos ) =
B
,
where is the angle between the photon and the electron
velocity and
B
= eB/m
e
c. When is not small, the
resonance condition gives

sc
10
3
B
14
_

1 keV
_
1
. (1)
3
The scattered photons are boosted in energy by the fac-
tor of
2
sc
. Such high-energy photons quickly convert
to e

pairs in the strong magnetic eld, creating more

particles near the star. A similar process of e

creation
operates in the polar-cap discharge of ordinary pulsars,
but in a dierent mode. In ordinary pulsars, the high-
energy photons convert to e

with a signicant delay.

The scattered photon initially moves nearly parallel to B
and converts to e

only when it propagates a sucient

distance where its angle

with respect to B increases

so that the threshold condition for conversion is satised.
This delay leads to the large unscreened voltage in pulsar
models.
In contrast, the magnetic eld of magnetars is so strong
that pair creation can occur immediately following reso-
nant scattering (Beloborodov & Thompson 2007). The
energy of a resonantly scattered photon is related to its
emission angle

by
E(

) =
E
B
sin
2

_
1
_
cos
2

+
m
2
e
c
4
E
2
B
sin
2

_
1/2
_
,
(2)
where E
B
= (2B/B
Q
+1)
1/2
m
e
c
2
is the energy of the rst
excited Landau level and B
Q
= m
2
e
c
3
/e 4.4 10
13
G.
The scattered photon may immediately be above the
threshold for conversion, E > E
thr
= 2m
e
c
2
/ sin

, if
B > 4B
Q
. Therefore, e

discharge in magnetars can

screen E

more eciently than in ordinary pulsars and

buer the voltage growth once the Lorentz factors of ac-
celerated particles reach
sc
given in Equation (1).
2.1. Pair creation on eld lines with apexes
R
max
<

2R
The discharge on twisted closed eld lines can be ex-
plored using a direct numerical experiment where plasma
is represented by a large number of individual particles in
the self-consistent electric eld. The existing numerical
simulations (Beloborodov & Thompson 2007) describe
the discharge on eld lines that extend to a moderate
radius R
max
<

2R, where R is the radius of the neutron

star. The magnetic eld is ultrastrong everywhere along
such eld lines, B B
Q
, and resonant scattering events
may eectively be treated as events of pair creation
a signicant fraction of scattered photons immediately
convert to e

.
The simulations demonstrate that voltage and pair cre-
ation self-organize in the twisted magnetosphere so that
a particle on average scatters 1 photon as it travels
through the electric circuit, maintaining the near-critical
multiplicity of pair creation M 1. This criticality con-
dition regulates the induced voltage to
e
10
9
V, which
accelerates e

particles to Lorentz factors 10

3
. The
electric circuit operates as a global discharge, in the sense
that the accelerating voltage is distributed along the en-
tire eld line between its footpoints on the star. It is
quite dierent from the localized gap that is usually
considered above polar caps in pulsars.
The discharge uctuates on the light-crossing timescale
R/c and persists in the state of self-organized critical-
ity. The behavior of the global circuit resembles a con-
tinually repeating lightning: voltage between the foot-
points of the eld line quasi-periodically builds up and
discharges through the enhanced production of charges.
M >> 1
M ~ 1
+

+
+

Fig. 2. Schematic picture of plasma circulation in the mag-

netosphere with surface B 10
15
G. Two regions are indicated.
(1) Inner corona. Here e

have a moderate multiplicity M 1.

The particles do not stop in the equatorial plane. The electric
eld E

ensures that electrons and positrons circulate in the op-

posite directions along the magnetic eld lines, maintaining the
electric current demanded by B. The particles are lost as
they reach the footpoints of the eld line and continually replen-
ished by pair creation. (2) Outer corona extended eld lines
with Rmax R. Electrons and positrons are created by the dis-
charge near the star and some of them ow outward to the region
of weaker B. Here resonant scattering enhances the pair multiplic-
ity, M 1, and decelerates the outow. The e

particles stop
at the apexes of magnetic eld lines (blue region in the equato-
rial plane), accumulate, and annihilate there. The number uxes
of electrons and positrons toward the annihilation region dier by
a small fraction M
1
, so that the outow carries the required
electric current j = (c/4) B. Electrodynamics of the twist
dissipation implies that the inner corona is less likely to be ac-
tive, as the electric currents are erased by the expanding cavity
(Figure 1); the observed activity tends to concentrate on extended
eld lines that form the outer corona.
The average plasma density in the circuit n is close to
the minimum density n
min
= j/ec, as required by the
criticality condition M= n/n
min
1.
The global-discharge picture applies only to eld lines
with R
max
<

2R and becomes irrelevant when the cur-

rents are erased in the inner magnetosphere as shown
in Figure 1. Then the observed activity must be asso-
ciated with currents on eld lines with large R
max
, i.e.
extending far from the star.
2.2. Pair creation on eld lines with apexes R
max
R
The discharge on extended eld lines may be expected
to have a similar threshold voltage
e
10
9
V, be-
cause the conversion of upscattered photons to e

is ef-
cient near the eld-line footpoints where B B
Q
. In
this zone, particles are able to resonantly scatter soft X-
rays once they are accelerated to 10
3
(Equation 1),
which requires
e
10
9
V. Further growth of voltage
would cause excessive creation of e

moving in both di-

rections, toward and away from the star, leading to e-
cient screening of E

.
A large fraction of the created particles must outow
to r R along the extended eld lines. The rate of
resonant scattering by a relativistic particle increases as
it moves from B B
Q
to B
<

B
Q
. The particle scat-
ters many more photons, because the resonance condition
shifts toward photons of lower energy
res
B whose
number density is larger. Note also that the eective
4
cross section for resonant scattering
res
= 2
2
r
e
c/
res
increases as B
1
(here r
e
= e
2
/m
e
c
2
is the classical elec-
tron radius). Practically all photons scattered by the
outowing particles in the region B
>

10
13
G convert
to e

; detailed Monte-Carlo simulations of this process

are presented in the accompanying paper (Beloborodov
2013). In essence, the particles outowing from the dis-
charge zone lose energy to photon scattering, and this en-
ergy is transformed to new generations of e

. As a result,
the e

multiplicity of the outow increases from M 1

to M 100. This implies that there is no charge starva-
tion in the outer corona there are plenty of charges to
conduct the current demanded by the twisted magnetic
eld.
Pair creation sharply ends near the surface of B
10
13
G; outside this surface the resonantly scattered pho-
tons are not absorbed. The steady relativistic outow
without pair creation maintains M = const along the
magnetic eld lines. This follows from conservation of
magnetic ux, charge, and particle number, which give
n/B = const, j/B = const, and n/j = const along the
eld line.
2.3. Global circulation of pair plasma
The picture of plasma circulation in the magnetar mag-
netosphere is summarized in Figure 2. The inner corona
(eld lines with R
max
of a few neutron-star radii) is
lled with the ultra-relativistic counter-streaming e

and
e
+
. In this region, resonant scattering is marginally e-
cient and a global discharge operates as described in Sec-
tion 2.1, with multiplicity M 1. Outside this region,
pair multiplicity is much higher and the electric current is
organized with both e

and e
+
outowing from the star.
The exact location of the boundary between the two re-
gions depends on the strength of the magnetic eld of
the star.
In the outer corona, the opposite ows in the nothern
and southern hemispheres meet in the equatorial plane of
the magnetic dipole and stop there. Two eects prevent
their inter-penetration. (1) Radiative drag is strong in
the outer corona and pushes both nothern and southern
ows toward the equatorial plane (see Section 3 below).
(2) When the two opposite ows try to penetrate each
other, a two-stream instability develops. As a result,
a strong Langmuir turbulence is generated, which in-
hibits the penetration. This eect is particularly impor-
tant in the transition region between the outer and inner
corona. For these eld lines, resonant scattering is e-
cient enough to generate M > 1 but not strong enough
to stop the pair plasma in the equatorial plane. Then the
colliding northern and southern ows are stopped by the
two-stream instability. This behavior contrasts with the
inner corona where the induced electric eld enforces the
counter-streaming of e

and e
+
, i.e. the opposite ows
with M 1 are forced to penetrate each other despite
the two-stream instability.
The density of e

pairs accumulated in the outer equa-

torial region (shown in blue in Fig. 2) is regulated by the
annihilation balance. In a steady state, the annihilation
rate is given by

N
ann
2M(I/e), where I is the electric
current through the annihilation region. The correspond-
ing annihilation luminosity is L
ann
= 2m
e
c
2
MI/e.
3. RADIATIVELY LOCKED CORONAL FLOW
Dynamics of the e

ow in the outer corona (r R) is

inuenced by resonant scattering, which exerts a strong
force F on the particles along the magnetic eld lines. F
vanishes only if the particle has the saturation momen-
tum p

such that the radiation ux measured in the rest

frame of the particle is perpendicular to B. In the sim-
plest case of a weakly twisted dipole magnetosphere ex-
posed to central radiation p

is given by (Appendix B),

2
p

(r, ) =
2 cos
sin
, (3)
where momentum is in units of m
e
c. The radiative force
always pushes the particle toward p = p

. The strength
of this eect may be measured by the dimensionless drag
coecient,
D
rF
p m
e
c
2
. (4)
Momentum p

is a strong attractor in the sense that

deviations p p

generate D 1 in the outer corona

(see Appendix A).
Even an extremely strong radiative drag does not
imply that e
+
and e

acquire exactly equal velocities

+
=

= p

(1 + p
2

)
1/2
. Such a single-uid
ow would be unable to carry the required electric cur-
rent j, and E

must be induced to ensure a sucient

velocity separation
+

. Below we describe the two-

uid model with
+
=

.
3.1. Two-uid model: basic equations
Consider an e

ow in the region where no new pairs

are produced. In a steady state, the uxes of electrons
and positrons are conserved (n

) = 0. This implies
j

= 0 where j

= en

are the contributions

of electrons and positrons to the net electric current j =
j
+
+ j

. Since e

move along the magnetic eld lines,

j

B where

are scalar functions. From j

=
0 and B = 0 one gets B

= 0. Therefore,
j
+
B
= const,
j

B
= const, (5)
are constant along the eld line. The net current j =
j
+
+j

is xed by the condition j = (c/4)| B|. The

multiplicity of pairs is dened by
M=
j
+
+|j

|
j
. (6)
Here + corresponds to positrons, which carry current
j
+
> 0 and corresponds to electrons, which carry
j

< 0; we assume a positive net current j = j

+
+ j

for deniteness. A counter-streaming model (v

+
v

< 0)
would have M = 1. A charge-separated outow (n

=
0) would also have M = 1. We study here pair-rich
outows with M> 1.
The outowing e

plasma must be nearly neutral,

3
n
+
n

, (7)
2
This expression is valid in the region where 1Br/B > (R/r)
2
.
In this region, stellar radiation may be approximated as a central
ow of photons, neglecting the angular size of the star R/r.
3
Here we neglect rotation of the neutron star and its magne-
tosphere, which is a good approximation everywhere except the
open eld-line bundle that connects the star to the light cylin-
5
otherwise a huge electric eld would be generated that
would restore neutrality. Using this condition, one nds
that en
+
v
+
en

= j is satised if
1

+
=
2
M+ 1
. (8)
A deviation from this condition implies a mismatch be-
tween the conduction current j
+
+j

and (c/4) B,
which would induce a growing electric eld according to
Maxwell equation E/t = c B 4 j. An electric
eld E

must be established in the outow to sustain

the condition (8) against the radiative drag that tends to
equalize

and
+
at

. This moderate electric eld

must be self-consistently generated by a small deviation
from neutrality, n = n
+
n

.
The two-uid dynamics of the outow is governed by
two equations,
m
e
c
2
d

dl
= F(

) eE

, (9)
where l is length measured along the magnetic eld line.
In the region of strong drag, |D| 1, the left-hand side
is small compared with F; then the radiative and electric
forces on the right-hand side nearly balance each other,
F(
+
) eE

and F(

) eE

. This implies,
F(
+
) F(

) (|D| 1). (10)

Equations (8) and (10) describe the radiatively locked
two-uid current with pair multiplicity M. The solution

to these equations exists if M > 1. Note that

<

<
+
in the radiatively-locked state.
Let us now relax the assumption |D| 1. Then
the inertial term m
e
c
2
d

/dl must be retained in Equa-

tions (9), i.e. we now deal with dierential equations
for

. Since

and
+
are not independent they
are related by condition (8) it is sucient to solve
one dynamic equation, e.g. for
+
(and use the dynamic
equation for

to exclude E

). Straightforward algebra
gives,
m
e
c
2
d
+
dl
=
F(
+
) +F(

)
1 + d

/d
+
, (11)
d

d
+
=
_
M1
M+ 1
_
2
_

+
_
3
. (12)
3.2. Sample numerical model
Suppose that e

plasma is injected near the star with

a given multiplicity, e.g. M = 50, and a given high
Lorentz factor, e.g.
+
= 100. The corresponding

is
determined by Equation (8). Suppose that the plasma
is illuminated by the blackbody radiation of the star of
temperature kT = 0.5 keV and neglect radiation from
the magnetosphere itself. This approximation is valid
der. In a more exact model, the Gauss law in the co-rotating
frame E = 4( + v) includes the eective vacuum charge
density v = B/2c where is the angular velocity of the
star (Goldreich & Julian 1969). Then the neutrality condition be-
comes e(n
+
n

) + v = 0. Magnetars rotate slowly (typical

1 rad/s) and hence a small fraction of their magnetic ux is
open, typically
<

10
4
. In the main, closed magnetosphere, the
condition |v/e| |j/ec| < n

is satised, and neutrality requires

n
+
n

with a high accuracy.

only for optically thin magnetospheres, which are con-
sidered here for simplicity. A more detailed model will
be developed in Sections 5 and 6, which will take into
account radiation scattered in the magnetosphere; then
radiation eld is not central, and we will have to solve
radiative transfer to determine the ow momentum.
An explicit expression for F() exerted by the central
radiation is given in Appendix B (Equation B6). The
steady-state solution for

in the outer corona can be

found by integrating Equation (11) along the magnetic
eld lines. The result is shown in Figure 3. The rel-
ativistic outow is injected near the star and initially
weakly interacts with the radiation; then it enters the
drag-dominated region |D| 1. The solution is not
sensitive to the precise radius of e

injection as long as
it is small enough, before the plasma enters the drag-
dominated region.
The electric eld in the region of |D| 1 is given
by eE

F(
+
), and the corresponding longitudinal
voltage established in the outer corona is found by inte-
grating F(
+
) along the eld line, e
e

_
F(
+
) dl.
Its typical value for the model in Figure 3 is 10
7
V.
Flows with lower Mdevelop stronger electric elds, how-
ever in all cases of interest (M 1) the drag-induced
voltage is below 10
9
V.
The calculations shown in Figure 3 assume that the
magnetospheric plasma is everywhere optically thin.
This is not so for real magnetars. Thompson et al.
(2002) showed that the characteristic optical depth of a
strongly twisted magnetosphere (twist amplitude 1)
is comparable to unity. When the large pair multiplicity
M is taken into account, the estimate changes to

M

1. (13)
This estimate describes the optical depth seen by pho-
tons that can be resonantly scattered by the ow, i.e.
the resonance condition (1 cos ) =
B
is satis-
ed somewhere along the photon trajectory. The large
implies the presence of scattered radiation in the mag-
netosphere, which is quasi-isotropic rather than central.
This increases the drag exerted on the outow and re-
duces p

. One may also expect a self-shielding eect: the

drag force F experienced by an electron (or positron)
is reduced by the factor of
1
. The problem of self-
consistent outow dynamics will be solved in Sections 5
and 6. The main feature seen in Figure 3 will persist in
the full self-consistent solution: the outow is strongly
decelerated (and drag-dominated) in the equatorial re-
gion at r (10 20)R.
4. TWO-STREAM INSTABILITY, ANOMALOUS
RESISTIVITY, AND RADIO EMISSION
4.1. Two-stream instability
The two-uid ow with v
+
> v

is prone to the two-

stream instability (e.g. Krall & Trivelpiece 1973). The
growth rate of the instability is obtained from the dis-
persion relation for Langmuir modes with frequency
and wavevector k, which can be derived by considering
perturbations

of the two-uid system and using con-

6
Fig. 3. Lorentz factors
+
(left panel) and

(right panel) in the two-uid model of the e

ow. In this example, the electric current

is carried by the e

outow of a xed multiplicity M = 50. The plasma is injected at radius r = 2R and outows along the magnetic
eld lines (white curves). The ow is illuminated by the star with temperature kT = 0.5 keV (magenta circle at the origin), and the
radiation exerts the forces F(

) on the positron (+) and electron () uids. The Lorentz factors

+
and

change as the ow enters

the drag-dominated region |D| 1. The region |D| > 3 is shown by the thick black curve and shadowed in black. D < 0 for positrons
(
+
> ) and D > 0 for electrons (

< ). The radiative drag stops the plasma in the equatorial plane outside 8R. A nearly dipole
magnetic eld (weakly twisted) with B
pole
= 10
15
G is assumed in this example. R is the neutron-star radius.
tinuity, Euler, and Poisson equations,
1

2
+

3
+
( kv
+
)
2


2

( kv

)
2
= 0, (14)
where
2

= 4n

e
2
/m
e
. The plasma is nearly neutral,
n
+
= n

, and hence
+
=

. Equation (14) has a so-

lution (k) with a positive imaginary part that describes
an unstable mode. The solution simplies in the follow-
ing two limits:
(1)
+

, which is valid when M

2

(see Equation 8). Then the ow is convenient to view

in its center-of-momentum frame that moves with
(
+
+

)/2. In this frame, the two uids with densi-

ties n

=
1
n

move in the opposite directions with

non-relativistic velocities v

= v, so the dispersion re-

lation (14) simplies. It gives the most unstable mode

k = (

3/2)

/ v with a growth rate

=

/2.
(2)
+
/

1. Then the contribution of positrons to

the dispersion relation is small compared to that of the
electrons. The growth rate of the instability is given by

1/2

1
+

p
(e.g. Lyubarsky & Petrova 2000).
This estimate gives the characteristic length-scale of
the instability c/, which is much shorter than the elec-
tron free path to resonant scattering,
sc
. Hence the ra-
diation drag and the induced electric eld eE

= F(p

)
are unable to lock the positive and negative charges
at the momenta p
+
and p

calculated in the two-uid

model. The instability will generate plasma oscillations
that should broaden the momentum distribution so that
particles ll the region p

< p < p
+
.
The generated plasma oscillations may be expected to
introduce an anomalous resistitivity. The uctuating E

in the oscillations creates a stochastic force that tends

to reduce the free-path of a charged particle. A simplest
estimate suggests that this eect could be very strong.
Suppose a substantial fraction of the energy density of
the ow m
e
c
2
n is given to plasma oscillations. Then the
characteristic electric eld is E

(8m
e
c
2
n)
1/2
; it is
much stronger than E

in the radiatively locked two-uid

model, however it is irregular and quickly changes sign.
Suppose that the stochastic electric force exerted on the
particle randomly changes sign on a timescale t
1
p
,
where
p
is the plasma frequency. The stochastic E

gives the momentum kicks P eE

1
p
and causes
diusion of particles in the momentum space with the
diusion coecient
D
p

(P)
2
t

(eE

)
2

p
. (15)
Diusion in momentum space p
2
(t) D
p
t implies a small
free-path of the particle, ()
2
m
2
e
c
3

p
/(eE

)
2
, much
smaller than the mean free path to resonant scattering.
Thus, a large anomalous resistivity could, in principle,
be possible, and then a large longitudinal voltage would
be generated to maintain the electric current.
4.2. Numerical experiment
To explore the role of the two-stream instability and
anomalous resistivity, we designed the following numer-
ical experiment. Keeping in mind that particles around
magnetars can ow only along the magnetic eld lines,
consider the simple one-dimensional problem. Suppose
7
Fig. 4. Snapshot of electric eld in the simulation box.
an e

beam is continually injected at the boundary z = 0

of the computational box 0 < z < L. The rate of electron
injection is smaller than the rate of positron injection, so
that the ow carries current j > 0. Positrons are injected
with xed p
+
and electrons with xed p

< p
+
; we chose
p
+
= 7 and p

= 0.5. The simulation keeps track of the

ow in the computational box of length L 150c/
p
.
The escape boundary condition is implemented at z = L.
The ow is simulated as a large collection of individual
particles (N 10
6
) that move in their collective electric
eld (see Beloborodov & Thompson [2007] for details of
the numerical method). The simulation was run for a
time of t 100L/c, long enough to see the quasi-steady
behavior.
The results of the simulation are as follows. As ex-
pected, strong plasma oscillations develop and persist in
the ow, as the instability is continually fed by the in-
jection of the two streams at z = 0. The uctuating
electric eld reaches very high amplitudes that could re-
verse a particle with Lorentz factor
+
on a short scale,
much shorter than L. A snapshot of E

(z) is shown in
Figure 4. The integral of the electric eld over z de-
termines the voltage
e
between the two boundaries of
the computational box. The measured
e
in the simu-
lation uctuates in time (Figure 5), and we calculated
its value averaged over 100L/c. This value turns out to
be very small, e
e
0.1
+
m
e
c
2
. Thus, the measured
anomalous resistivity is small, in sharp contrast with the
simplest estimate (15) that would predict L and
hence e
e

+
m
e
c
2
.
The failure of the estimate (15) is related to the as-
sumption that the stochastic electric force applied to
the particle is random, uncorrelated on timescales longer
than
1
p
. The numerical simulation indicates that this
assumption is incorrect. Apparently, a complicated time-
dependent pattern is organized in the phase space, which
allows the charges to nd small-resistance paths through
the waves of E

and conduct the current at a low net

voltage and a low dissipation rate.
One limitation of the numerical experiment should be
noted: the one-dimensional computational box allows no
dependence of E

on the transverse coordinates x and

y, which excludes the coupling of Langmuir waves to the
transverse electro-magnetic modes. A two-dimensional
(or a complete three-dimensional) simulation will be
needed to explore the role of the transverse modes. We
anticipate that a low anomalous resistivity will be found
Fig. 5. Voltage across the simulation box as a function of time.
Fig. 6. Snapshot of the momentum distributions of e
+
(red)
and e

(blue). Vertical lines show the injection p

at z = 0.
in the full simulations. A high resistivity would imply
a quick dissipation of magnetospheric currents, which
would produce a high luminosity and quickly erase the
magnetic twist. This is not supported by observations
of magnetars. The observed luminosities and evolution
timescales are consistent with the model neglecting the
anomalous resistivity, where voltage is controlled by the
threshold of the e

discharge, e
e
10
9
10
10
V.
The numerical simulation shows that the two-stream
instability signicantly changes the momentum distribu-
tions of electrons and positrons from the injected delta-
functions (p p

) and (p p
+
). The e

distributions
are broadened so that they ll the region between the
injection momenta p
+
and p

(Figure 6).
4.3. Low-frequency emission
8
An important implication of the two-stream instabil-
ity is the excitation of a strong plasma turbulence that
can generate coherent low-frequency radiation. The two-
stream instability is often considered in pulsar models
as a mechanism feeding radio emission from the open
eld-line bundle (e.g. Sturrock 1971; Cheng & Rud-
erman 1977). A related model invoking radiative drag
was considered by Lyubarsky & Petrova (2000).
4
In the
e

ows around magnetars, the two stream instability

is naturally driven by the strong electric current in the
system that tends to lock itself in the two-uid congura-
tion as described in Section 3. The instability is continu-
ally pumped by the radiative drag in the dense radiation
eld of the magnetar. The generated low-frequency ra-
diation has the best chance to escape near the magnetic
axis, where the plasma density is lowest and its Lorentz
factor is highest. Note that the two-stream instability
operates on the closed (twisted) eld lines, which carry
signicant magnetic ux. This allows the low-frequency
emission to be unusually bright, even though the volt-
age
e
10
9
10
10
V is small by the ordinary-pulsar
standards.
Radio pulsations have been detected and studied in de-
tail in two magnetars XTE 1810-197 and 1E 1547.0-5408
(Camilo et al. 2006; 2007). The estimated radio luminos-
ity L
r
10
30
erg s
1
requires a suciently high ohmic
dissipation rate, I
e
> L
r
, which cannot be generated
in the open eld-line bundle unless
e
exceeds L
r
/I
LC

10
11

1
r
L
r,30
V. Here
r
is the eciency of radio emis-
sion, I
LC
is the electric current circulating in the open
bundle, I
LC
c /R
2
LC
, and R
LC
= cP/2 10
10
cm is
the light cylinder for a magnetar rotating with the pe-
riod P of a few seconds. As we argued above, voltage
is likely near the threshold of e

discharge, which gives

e
10
11
V.
Therefore, we conclude that the electric current asso-
ciated with observed radio emission is large, I I
LC
,
and should ow in the closed (twisted) magnetosphere,
giving a bright and relatively broad radio beam. This
is consistent with the unusually broad radio pulses of
magnetars, much broader than the typical pulse of ordi-
nary pulsars with similar periods (Camilo et al. 2006;
2007). Note also that the plasma density in the twisted
closed magnetosphere reaches much higher values than
in the open eld-line bundle, and the plasma frequency
may approach the infrared band. This may explain the
observed hard radio spectra.
One could consider the possibility that the radio lumi-
nosity of magnetars is generated by enhanced dissipation
in the open eld-line bundle and its immediate vicin-
ity. Thompson (2008b) suggested that the diusion of
magnetic twist in the closed magnetosphere initiates a
strong Alfvenic turbulence near the light cylinder, with
a high dissipation rate. This picture assumes that the
magnetic twist tends to spread due to ohmic diusion, as
observed in normal laboratory plasma with a nite resis-
4
They suggested that a broad momentum distribution of rel-
ativistic particles in the open eld-line bundle can evolve into a
two-hump distribution as a result of resonant scattering losses, as
the lower energy particles lose their energy faster than the more
energetic ones. In contrast, the two-uid ow described in Sec-
tion 3 is shaped by the induced E

so that it sustains the electric

current j; without E

radiative drag would equalize the velocities

of all particles at v.
tivity. However, later work (Beloborodov 2009; 2011a)
showed that the twists in neutron-star magnetospheres
evolve dierently: the twist is erased inside out rather
than spreads diusively. The twist evolution /t is
controlled by voltage induced along the magnetic eld
lines,
e
, so that /t d
e
/d, where is the mag-
netic ux function labeling the eld lines ( = 0 on
the magnetic dipole axis). An increased voltage near the
axis would imply a large negative d
e
/d, which implies
a large negative /t, i.e. rapid untwisting. Thus, the
high-
e
twist near the open eld-line bundle cannot be
sustained. The ohmic eects in the closed magnetosphere
can pump the twist near the open bundle only if
e
is re-
duced toward the magnetic dipole axis, i.e. d
e
/d > 0.
Then the twist pumping continues until the expanding
cavity j = 0 reaches the open bundle; a snapshot of
this evolution is shown in Figure 1. The pumping of
leads to outbursts and spindown anomalies (Parfrey et
al. 2012, 2013).
Our scenario for the low-frequency emission from mag-
netars may be summarized as follows. The emission is
generated by the two-stream instability on the twisted
closed eld lines with the apex radii R
max
such that
R R
max
R
LC
. These eld lines carry a large elec-
tric current I c /R
2
max
I
LC
and a modest voltage

e
10
9
10
10
V. The high plasma density and the
broad beam of radiation expected on these eld lines ex-
plain the unusual radio pulsations of magnetars.
5. DYNAMICS OF OUTFLOW WITH A BROAD
MOMENTUM DISTRIBUTION
5.1. Waterbag model
The plasma instability discussed in Section 4 is gener-
ated by the gradient of the distribution function df
e
/dp,
and the feedback of the excited plasma waves tends to
make the distribution atter. The numerical simulation
in Section 4 illustrates how two interpenetrating cold u-
ids of e
+
and e

with f
e
(p) = (1/2)[(pp

)+(pp
+
)]
quickly evolve into a state with a broad and smooth
f
e
(p). Below we design a simple modication of the two-
uid model that takes this eect into account.
The simplest model has a top-hat distribution func-
tion. In plasma physics, this approximation is often
called waterbag model. Like the two-uid model, the
outow is described by two parameters p

(or

). How-
ever, now instead of two delta-functions we require the
e

distribution to be at between p

and p
+
,
f
e
(p) =
_
(p
+
p

)
1
p

< p < p
+
0 p < p

or p > p
+
(16)
This distribution includes both electrons and positrons.
The plasma must be nearly neutral, n
+
= n

, and the
total density of particles n = n
+
+ n

is given by
n =

c
=
Mj
e

c
, (17)
where

N = Mj/e is the particle number ux and

is
the average velocity,

=
_
(p) f
e
(p) dp =

+

p
+
p

. (18)
9
Fig. 7. Waterbag distribution of minimum width.
Similar to the two-uid model, p
+
and p

are not inde-

pendent, as the outow must carry current j with a given
multiplicity M; this would be impossible if, for instance,
p
+
= p

. The minimum possible width of the distribu-

tion function (i.e. minimum p
+
p

) is achieved if all
negative charges are slower than all positive discharges,
as shown in Figure 7. We will adopt this idealized mo-
mentum distribution in our numerical simulations. This
is a rather crude approximation to more realistic distri-
butions of e
+
and e

, which overlap in momentum space

(cf. Figure 6).
5
Nevertheless, it is useful as it allows one
to explore all basic features of the e

outow using a
concrete relation between p
+
and p

. This relation is
determined by the outow multiplicity M. It is easy to
show that the generalization of the two-uid Equation (8)
to e

ows with any distribution function is given by

1

+
=
2
M+ 1
, (19)
where

+
and

are the average velocities of the posi-

tive and negative charges, respectively. For the waterbag
distribution shown in Figure 7 this condition gives

+
=
( p)

+
( p)
= 1
2
M+ 1
, p =
p

+ p
+
2
, (20)
where ( p) = (1 + p
2
)
1/2
. Equation (20) determines the
relation between p
+
and p

(for a given M); it is shown

in Figure 8.
Taking into account the relation between p
+
and p

,
the outow has essentially one degree of freedom besides
5
The minimum-width waterbag model may be particularly
crude near the star where the outow just begins to experience
signicant radiative drag. The faster positive charges will experi-
ence drag rst while the slower negative charges still move freely.
As p
+
decreases, the minimum-width model would require an in-
crease in p

. In reality, p

may not react to the reduced p

+
until
particles with p = p

also begin to experience the drag force (which

is positive as p

< p). After this point, the outow may not be

far from the minimum-width waterbag state.
Fig. 8. Relation between p
+
and p

for the waterbag model

shown in Figure 7. Each curve describes an outow of a given
multiplicity M, which is indicated next to the curve; p

= 0 is
indicated by the dotted line.
the multiplicity M. One can chose, e.g., p
+
as an inde-
pendent variable, or any convenient combination of p
+
and p

, (p
+
, p

), that is independent from M(p

+
, p

).
Below we will choose variable and formulate the dy-
namic equation for ; it will describe how the interaction
with radiation governs the outow dynamics along the
magnetic eld lines.
5.2. Momentum equation
Since the outow is bound to move along the mag-
netic eld lines, we will need the projection of momentum
equation onto B. It is convenient to write this equation
in a covariant form that is valid in any curved coordinate
system x
i
(e.g. spherical coordinates). In a steady state,
the momentum equation reads
B
i
B

k
T
ik
=
dP
dt dV
, (21)
where dP/dtdV is the rate of momentum exchange with
radiation (per unit volume),
k
is the covariant deriva-
tive (indices i, k run from 1 to 3), and T
ik
are the com-
ponents of the plasma stress tensor,
T
ik
= m
e
c
2
_
u
i
u
k
dn

= nm
e
c
2
B
i
B
k
B
2
_
p
2

f
e
(p) dp.
(22)
Here u
i
= p B
i
/B are the spatial components of the four-
velocity vector of a particle with dimensionless momen-
tum p, dn = nf
e
(p)dp is the number density of particles
with momenta (p, p + dp), and dn/ is the correspond-
ing density in the frame moving with (p) = (1 +p
2
)
1/2
.
Then the left-hand side of Equation (21) takes the form,
B
i
B

k
T
ik
=
B
i
B

k
_
nm
e
c
2
B
i
B
k
B
2
p
_
=m
e
c
2
B
k

k
_
n
B
p
_
, (23)
10
where we used B
i

k
(B
i
B
k
/B) = 0 (which follows from

k
B
k
= 0). The directional derivative B
k

k
equals
B d/dl where l is length measured along the magnetic
eld line. Equation (23) can be further simplied using
n

/BM= const, which follows from the relation

n = M
j
ec

, (24)
and j/B = const (cf. Equation 5). This gives
B
i
B

k
T
ik
= m
e
c
2
n

M
d
dl
_
M

p
_
. (25)
When M = const (i.e. no new pairs are created), M
cancels out. Finally, the substitution of Equation (25)
to Equation (21) gives the momentum equation in the
following form,
d
dl
=

m
e
c
2
,
p

, (26)
where

F = n
1
dP/dtdV is the average force exerted by
radiation per particle. If

F = 0 then remains constant
along the eld lines, which implies p

= const the mo-

mentum distribution remains unchanged along the ow.
For the waterbag model,

is given by Equation (18).
The quantity p can also be expressed in terms of p

,
using the indenite integral
_
p dp =
_
p d =
p
2

1
4
ln
_
1 +
1
_
+ const. (27)
This gives
=
1

_
1
2
(p
+

+
p

)
1
4
ln
(1 +
+
)(1

)
(1
+
)(1 +

)
_
.
(28)
In our numerical simulations, we use as the variable
that describes the dynamical state of the outow, and
solve the dierential Equation (26) for . The values of
p

that correspond to a given and M are found from

Equations (20) and (28).
The force

F experienced by an outow with a given is
determined by the local radiation eld, which is described
by the intensities in the two polarization modes I

(, n)
and I

(, n) (Appendix A). In general, the force acting

on a plasma with a distribution function f
e
(p) is given
by Equation (A20). Force

F is easily calculated in the
simple case of an optically thin magnetosphere exposed
to the central blackbody radiation (Appendix B). Then
the right-hand side of Equation (26) is a known function
of and it is straightforward to numerically solve this
equation; the results are similar to the two-uid model
described in Section 4. The optically thin approximation
is, however, invalid for active magnetars. The radiation
is not central; instead I

and I

must be calculated self-

consistently, together with the outow dynamics. This
requires radiative transfer simulations.
Note that we assume here that p

> 0, so that all

particles move away from the star until they reach the
equatorial plane and disappear (annihilate) there. This
approximation is reasonable for the ow along the ex-
tended eld lines with apex radii R
max
>

10R. Near
the apexes, i.e. near the equatorial plane = /2, the
radiative drag enforces p

1, creating a dense
layer of slow particles where they annihilate (Fig. 2). A
more general model of the ow could allow the parti-
cles to cross the equatorial plane and enter the opposite
hemisphere. By symmetry, this would be equivalent to
the reection boundary condition, i.e. the mirror image
of the outow approaching the equatorial plane would
emerge from the equatorial plane. Then the distribution
function f
e
(p) must extend to negative p. This modica-
tion would be required for the plasma ow on eld lines
with small R
max
, where radiative drag is less ecient and
the plasma can cross the equatorial plane with a large p.
In this paper, we focus on the eld lines with R
max
>

10,
where this does not happen. The eld lines with small
R
max
are assumed to form a cavity with j = 0 and a
negligible plasma density (Figure 1).
6. SELF-CONSISTENT RADIATIVE TRANSFER
The problem of radiative transfer in a relativistically
moving e

plasma whose velocity is controlled by the ra-

diation eld is not unique to magnetars. A similar situ-
ation may occur in accretion disk outows (Beloborodov
1998) and gamma-ray bursts (Beloborodov 2011b). The
strong radiative drag (measured by the coecient D 1,
Equation [4]) was previously shown to simplify the prob-
lem, as it forces the plasma to keep the saturation mo-
mentum p

such that the net radiation ux vanishes

in the plasma rest frame. This equilibrium transfer
has one additional integral compared with the classical
Chandrasekhar-Sobolev transfer problem for a medium
at rest. The transfer in magnetar magnetospheres has,
however, two special features that complicate the prob-
lem. First, the electric current and plasma instabili-
ties imply additional (electric) forces that broaden the
momentum distribution around p

, as discussed in Sec-
tions 4 and 5. Therefore, the equilibrium condition
p = p

is not satised even where D 1. Second, opac-

ity is dominated by resonant scattering, whose rate is
sensitive to the particle momentum. Below we develop
a method to solve the self-consistent transfer for magne-
tars.
6.1. Monte-Carlo technique and the virtual beam
method
Radiative transfer in a magnetosphere lled with
plasma with given parameters can be calculated using
the standard Monte-Carlo technique (e.g. Fern andez &
Thompson 2007; Nobili, Turolla & Zane 2008). Black-
body photons are injected into the magnetosphere at the
neutron-star surface and their trajectories are followed
until they escape the magnetosphere. We implement this
method using the scattering opacity given in Appendix A
and keeping track of the photon polarization, which can
switch in the scattering events.
Calculation of transfer in an outow with self-
consistent dynamics is a more ambitious goal, as the
plasma parameters are not known in advance. A nat-
ural approach is iterative. One can start with a trial
outow, calculate radiative transfer to nd the radiation
intensity that would correspond to this outow, and then
re-calculate the outow dynamics in the obtained radi-
ation eld. These iterations can be repeated until they
converge. The problem is simplied for the waterbag
11
plasma model (Section 5) as the outow is described by
one dynamic variable . As the rst trial initiating the
iterations one can take the outow solution (r, ) for
the optically thin magnetosphere exposed to the central
blackbody radiation.
One then encounters the following diculty. For the
calculation of next iteration of outow dynamics, one
needs to know I

(r, , n) and I

(r, , n) everywhere in
the magnetosphere. In axisymmetric magnetospheres,
the radiation intensity is a function of ve variables: two
for location (radius r and polar angle ), one for spec-
trum (), and two for angular distribution (unit vector n
is described by two angles). To determine intensities I

and I

with sucient accuracy, one has to introduce a

ve-dimensional grid and accumulate large photon statis-
tics for each grid cell during the Monte-Carlo simulation
of radiative transfer. A grid of size N for each of the ve
variables has N
5
cells. Accumulation of photon statistics
in each cell requires the calculation of a huge number of
Monte-Carlo realizations of the photon trajectory in the
magnetosphere. This is expensive and gives poor accu-
racy.
There is, however, a more ecient method that does
not require the knowledge of intensities I
,
(r, , n).
They are not, in fact, needed for the calculation of
outow dynamics. What enters the momentum Equa-
tion (26) is the average force

F(r, ), which may be tab-
ulated in the two-dimensional space of r, . This force
can be calculated directly during the Monte-Carlo sim-
ulation of radiative transfer if we nd a way to evaluate
the contribution of each simulated photon to

F(r, ) as
it propagates through the magnetosphere.
This can be achieved if we imagine that the photon is
replaced by a beam of radiation. As we follow the photon
along its trajectory, we can calculate the force that would
be applied by the imaginary beam to any given outow
that can be imagined in the magnetosphere. In the wa-
terbag model, the outow is described by one dynamical
parameter , and so the force applied by the beam can
be tabulated on a grid of .
Once we know how to calculate the force created by
each photon trajectory in our Monte-Carlo simulation,
we should be able to average it over all simulated pho-
tons and thus accurately evaluate

F(r, , ). This gives
the force that would be applied by our radiation eld
to an outow with any given at any point r, . At
the next iteration step

F(r, , ) is used to obtain the
new outow solution (r, ) by integrating the momen-
tum equation d/dl = W(l, ), where W =

F/

m
e
c
2
(Equation 26). Then the Monte-Carlo simulation can be
repeated to calculate radiative transfer in the new out-
ow and nd

F(r, , ) for the new radiation eld. These
steps can be repeated until the outow solution (r, )
converges, i.e. remains practically unchanged by new it-
erations.
The concrete implementation of this strategy is as fol-
lows. Let L
th
be the thermal luminosity of the star and

N = L
th
/2.7kT be the number of photons emitted by the
star per unit time. In our Monte-Carlo simulation we fol-
low K
MC
10
7
random photon trajectories. This can be
thought of as dividing

N into K
MC
random monochro-
matic beams. Each beam has a random start at the
star surface (the photon energy is drawn from the Planck
distribution) and follows one random realization of the
photon trajectory, which can involve multiple scattering
events in the magnetosphere. The photon number ux
in each monochromatic beam is

N
b
=

N/K
MC
, and the
energy ux in the beam is

E
b
=

N
b
. (29)
Note that

N
b
= const along the beam (i.e. along the
photon trajectory in the Monte-Carlo simulation) while
the photon energy changes after each scattering in the
magnetosphere. The collection of K
MC
beams represent
the state of the radiation eld around the star.
As we follow each realization of the photon trajectory,
in parallel we calculate the force applied by the virtual
beam to the plasma. We can imagine that the beam has
a small cross section A (it will cancel in the nal result)
and ux density F =

E
b
/A. Equation (A19) gives a
general expression for the force exerted by radiation on
a plasma with a given distribution function f
e
(p). For
a monochromatic beam of frequency propagating at
angle with respect to the magnetic eld, this expression
gives the following momentum deposition rate,
dP
dt dV
= 2
2
r
e
nF

B

2
[
1
f
e
(p
1
)
2
f
e
(p
2
)] , sin
B
,
(30)
(dP/dtdV = 0 if sin >
B
), where
p
1,2
(, ) =

B
sin
2

_
cos

1

2

2
B
sin
2

_
, (31)
and the factor depends on the beam polarization,
=
_
_
_
1,
1

2

2
B
sin
2
,
(32)
The rate of momentum deposition by the beam (i.e. the
exerted force) per unit length along its trajectory is given
by
dP
dt ds
=
dP
dt dV
A. (33)
We need to tabulate the force on a spatial grid, which
is used to calculate the outow dynamics at the next
iteration. Therefore, we need to evaluate the net force
applied to a given spatial cell. The number of particles
in the cell is nV
c
where V
c
is the cell volume, and the
force exerted by the beam per particle in the cell is given
by
d

F
b
ds
=
1
nV
c
dP
dt ds
= 2
2
r
e
V
c

B

E
b

2
[
1
f
e
(p
1
)
2
f
e
(p
2
)] .
(34)
In the transfer calculation, we track the photon trajec-
tory using small steps s, much smaller than the cells of
the spatial grid. To obtain the force

F
b
(r, , ) applied
by the beam in a given cell we integrate Equation (34)
along the photon path where it crosses the cell. Note
that a ner spatial grid implies a larger d

F
b
/ds V
1
c
;
however, it also implies that the cell is less frequently
visited by photons in our Monte-Carlo simulation, and
the photons spend a shorter time in the cell; therefore,
the nal result does not depend on the grid.
12
In an axisymmetric magnetosphere, the spatial cells
(i, j) are tori of volume V
i,j
= 2r
2
i
sin
j
r . The net
force applied per particle in a given cell (i, j) is obtained
by summing up the contributions

F
b
(r
i
,
j
, ) from all
simulated beams,

F(r
i
,
j
, ) =

N
K
MC
KMC

k=1
2r
e

V
i,j

_
cell(i,j)

[
1
f
e
(p
1
)
2
f
e
(p
2
)] ds. (35)
Note that the beam may cross a given cell multiple times
in an opaque magnetosphere, and all crossings contribute
to the path integral in Equation (35). The e

distri-
bution function f
e
(p) is determined by the parameter
(assuming a given pair multiplicity M, see Section 5).
As a test, one can apply Equation (35) to the simplest
case of a transparent outow exposed to the central ther-
mal radiation. Then the expected

F can be directly cal-
culated using Equation (B17) in Appendix B. We analyt-
ically veried that in this case Equation (35) is reduced
to Equation (B17). We also tested our numerical code;
it reproduced the analytical result.
6.2. Results
The obtained self-consistent solution for the outow
dynamics is shown in Figure 9. We show p
+
rather than
our dynamical parameter , because p
+
is closely related
to the radiation emitted by the outow. We nd that
only particles with the highest momenta p p
+
reso-
nantly scatter thermal photons in the relativistic zone
p
+
> 1; the remaining, dominant part of the momen-
tum distribution does not participate in scattering. The
Lorentz factor of the scattering particles is

sc
(1 + p
2
+
)
1/2
. (36)
The solution p
+
(r, ) shown in Figure 9 was calculated
assuming that the star has radius R = 10 km, a uni-
form surface temperature kT = 0.3 keV, and a moder-
ately twisted dipole magnetic eld with B
pole
= 10
15
G
and = 0.3; the multiplicity of the e

ow was xed
at M = 200. The ow is injected at r = 2R with
p
+
(2R) = 100. The choice of the boundary condition
is not important as we are interested in the ow be-
havior outside 5R, where the scattered photons avoid
conversion to e

pairs and can escape, i.e. where the

observed hard X-rays are produced (Beloborodov 2013).
For any reasonable boundary value p
+
(2R), the ow re-
laxes to the same solution p
+
(r, ) outside a few stellar
radii. Remarkably, our simulations gave practically the
same solution for a broad range of parameters , M, T
that is relevant for magnetars. This demonstrates a ro-
bust self-regulation mechanism; this important feature is
explained below (see also the accompanying paper).
There are two distinct zones in the outow:
I. Non-relativistic zone p
+
1, which is near the equa-
torial plane (red zone in Figure 9). This zone has a huge
optical depth and scatters essentially all thermal photons
that impinge on it from the star ( 10% of the thermal
luminosity L
th
). We will call this zone the equatorial
reector.
Fig. 9. Self-consistent outow solution. Color shows the pa-
rameter p
+
of the waterbag distribution function. Observed hard
X-rays (discussed in the accompanying paper Beloborodov (2013)),
are produced by particles with p p
+
. The simulation assumed
that the active j-bundle occupies the eld lines with apex radii
Rmax > 10R (Section 1).
II. Relativistic zone p
+
> 1 (blue to green in Figure 9).
This zone is transparent for essentially all thermal pho-
tons owing from the star, except for rare photons in the
far Wien tail of the Planck spectrum, which may be ne-
glected. Basically, the outow in this zone does not see
the thermal radiation owing from the star. It mainly
interacts with (and is decelerated by) the quasi-thermal
photons that ow from the equatorial equator.
It is easy to see why the outow in the relativistic
zone
+
1 is regulated so that it interacts with a
tiny fraction of the thermal photons around the magne-
tar. Scattering on average boosts the energy of a thermal
photon by a factor comparable to
2
, and hence the
energy lost by the outow per scattering is
2
. If
we imagine that each thermal photon is scattered once
with an average blueshift of
2
, the generated hard X-ray
luminosity
2
L
th
would exceed the kinetic power of the
outow L, which is impossible. The scattering rate must
be kept low, just sucient for the self-consistent grad-
ual deceleration of the outow. The outow vision
of targets for scattering is controlled by the resonance
condition (1 cos ) =
B
, where is the angle be-
tween the photon direction and the particle velocity

.
The scattering rate is greatly reduced if is so low that
the resonance condition is satised only for rare photons
in the Wien tail of the thermal spectrum. The situa-
tion may be compared with the regulation of the nuclear
burning rate in the sun. The self-consistent temperature
is suciently low so that fusion reactions occur only in
the far tail of the Maxwell distribution; as a result only
rare particles participate in burning, and these particles
13
are concentrated in a narrow interval of (large) momenta,
a phenomenon known as the Gamow peak. Similarly, our
self-consistent outow moves suciently slow so that res-
onant scattering is only enabled between rare particles
with the highest p p
+
and rare thermal photons with
the highest energies in the particle rest frame. These
lucky photons have the largest (gained after reection
from the equatorial reector) and large 2.7kT.
The inward direction (cos < 0) of the reected photons
gives them particularly large blueshift in the outow rest
frame and hence reduces the energy requirement in the
lab frame. This makes the reected photons the domi-
nant targets for scattering even though their density is
much smaller than the density of photons owing directly
from the star. The reected photons have a diluted quasi-
thermal spectrum of temperature T.
In essence, we observe in our simulations that the out-
ow moves fast enough to resonantly interact with re-
ected photons of energy (7 10)kT (the low-
density exponential tail of the thermal spectrum), and
slow enough to not interact with the main peak of the re-
ected thermal spectrum 3kT. This condition, to-
gether with the resonance condition (1 cos ) =
B
and cos 0.5, determines that the scattering plasma
moves with Lorentz factor

sc

m
e
c
2
10kT
B
B
Q
, (37)
as long as
sc
1. In this regime, the number of tar-
get photons visible to an outowing particle has a strong
exponential sensitivity to the particle momentum p. Es-
sentially all scattering must be done by particles with
the highest momenta in the distribution function, i.e.
with p p
+
for the waterbag distribution, as indeed
observed in our numerical simulation. Therefore,
sc
is
associated with p
+
. Equation (37) serves as a simple and
reasonably accurate approximation to the exact numer-
ical results shown in Figure 9. Its applicability is not
limited to the specic simulation with its B
pole
, T, ,
M the approximation works well for other magnetar
parameters, because of the robust self-regulation eect
described above. This fact is further discussed and illus-
trated in Figure 2 in Beloborodov (2013).
Besides Equation (37), the transfer problem is char-
acterized by the position of the equatorial reector. It
is described by a simple formula, which can be used to
scale the results shown in Figure 9 to models with other
parameters. The formula is based on the following fact
(see Appendix B): if a given active magnetic loop extends
to the region where
B
< 20kT, the central thermal ra-
diation exerts a suciently strong drag on the outow
to bring it to rest at the top of the loop. The region

B
< 20kT corresponds to r > R
1
where
R
1
80
_

10
33
G cm
3
_
1/3
_
kT
1 keV
_
1/3
km. (38)
Here = R
3
B
pole
/2 is the magnetic dipole moment of
the star. Equation (38) describes the position of the inner
edge of the non-relativistic (red) zone; e.g. the model in
Figure 9 has R
1
10R.
It is instructive to compare the result of the full trans-
fer calculation in Figure 9 with the simplest, optically
thin two-uid model in Figure 3. The relativistic zone
p
+
> 1 remains practically transparent to thermal radi-
ation. The key dierence is the presence of the opaque
equatorial reector. The reector weakly aects the
spectrum of thermal photons supplied by the star, how-
ever it signicantly changes their angular distribution
in the magnetosphere. As a result, the radiation exerts
a stronger drag on the outow and p
+
decreases faster
along the magnetic eld lines. In Figure 3,
+
p
+
1
remains huge near the magnetic axis the central ra-
diation is unable to decelerate the plasma because the
photons ow from behind and have small angles with
respect to the plasma velocity. In Figure 9, the equatorial
reector supplies photons with large , which eciently
decelerate the outow, according to Equation (37).
The upscattered photons of energy E
2
E
t
are
beamed along the relativistic outow. Therefore, they
become unable to decelerate the plasma, even though
they can scatter multiple times before escaping. The
outow signicantly loses energy when it scatters a pho-
ton propagating at a large angle with respect to the
outow velocity; only in this case the scattering boosts
the photon energy by the factor of
2
sc
1. After
the scattering, the photon angle is reduced to
1
sc
,
and its subsequent scatterings have a small eect on the
outow dynamics. The beamed radiation initially moves
together with the plasma and then escapes. Our transfer
simulations include all scattering events, however practi-
cally the same p
+
(r, ) would be obtained if only single
scattering were allowed in the relativistic zone p
+
1.
7. CONCLUSIONS
This paper examined the behavior of the relativis-
tic plasma created by e

discharge around magnetars.

Plasma circulation in the magnetosphere is schematically
shown in Figure 2. We focused on large magnetic loops,
which must be heavily loaded with e

pairs. The plasma

momentum is controlled by the radiation eld around
the star, which interacts with e
+
and e

via resonant
scattering. We developed a method to calculate radia-
tive transfer in the self-consistently moving plasma and
obtained the solution for the e

ow (Figure 9). The so-

lution is a strong attractor the behavior of the plasma
outside a few stellar radii does not depend on how the
plasma is injected near the star and its initial Lorentz
factor
0
, as long as
0
30 (
0
>

10
3
is expected,
which corresponds to the discharge voltage
e
10
9
V).
The e

ow shows the following features:

(1) The relativistic ow scatters radiation with a well
dened Lorentz factor
sc
given in Equation (37);
sc
decreases proportionally to B along the magnetic eld
lines.
(2) The relativistic owremains transparent to thermal
photons emitted by the star until it decelerates to non-
relativistic momenta p < 1. The non-relativistic zone is
opaque to the star radiation and forms the equatorial
reector (red zone in Figure 9).
(3) The energy lost by the decelerating ow is con-
verted to hard X-rays. The resulting emission is calcu-
lated and compared with observations in the accompa-
nying paper (Beloborodov 2013).
(4) The plasma is nearly neutral, n
+
n

, and carries
the electric current j = (c/4) B by adjusting the
particle velocities. In large magnetic loops (extending
14
to the region of B
<

10
13
G) the plasma has a high e

multiplicity M 10
2
, and both electrons and positrons
outow from the neutron star, with a small separation
in the velocity space. This separation is sustained by
a modest electric eld induced along the magnetic eld
lines.
(5) The enforced electric current and radiative drag to-
gether create a conguration that is prone to two-stream
instability, which is expected to generate low-frequency
radiation. The mechanism of initiating the two-stream
instability is unique to magnetars, explaining their spe-
cial radio emission and possibly optical/UV emission.
This work was supported by NASA grants NNX-10-
AI72G and NNX-13-AI34G. I thank R. Hascoet for com-
ments on the manuscript.
APPENDIX
A. RESONANT SCATTERING
Resonant scattering plays a signicant role in ordinary pulsars (e.g. Kardashev, Mitrofanov, & Novikov 1984;
Daugherty & Harding 1989; Sturner 1995; Lyubarsky & Petrova 2000). It is also the dominant radiative process in
magnetar magnetospheres, which governs the radiative transfer calculated in this paper. Below we summarize basics
of resonant scattering, write down the cross section, the optical depth of the e

ow, and the radiative drag force that

are used in our numerical simulations.
Photons scattered in the region B > 10
13
G are immediately absorbed (Beloborodov 2013. Interesting radiative
transfer occurs outside this region, where B B
Q
. In particular, the observed hard X-rays of energy up to several
MeV are radiated where B B
Q
. Electron recoil is small for resonant scattering in such relatively weak elds, and
the scattering cross section is particularly simple.
A.1. Scattering cross section
In classical language, the electro-magnetic wave (photon) with frequency
B
= eB/m
e
c resonates with the Larmor
rotation of electron. Then the wave strongly accelerates the charged particle and generates scattered radiation. The
corresponding cross section is largest for waves with the right-hand circular polarization e = e

that matches the

electron Larmor rotation. Here e

= 2
1/2
(e
x
ie
y
) and {e
x
, e
y
, e
z
} is a Cartesian basis with the z-axis anti-parallel
to B. For a wave with an arbitrary polarization vector e, only the projection of e on e

is responsible for the resonance,

and the cross section is reduced by the factor |e

|
2
, where e

is the complex conjugate of e. In quantum language,

the resonance occurs because the photon energy matches the energy
B
needed for the electron transition from the
ground Landau state to the rst excited state. The resonance has a nite width. It equals the natural width of the
cylcotron line , which corresponds to the lifetime of the excited electron to spontaneous transition back to the ground
Landau state.
For an electron at rest, the dierential cross section for photon scattering into solid angle d

is given by (Canuto
et al. 1971; Ventura 1979),
d
d

= r
2
e

2
(
B
)
2
+ (/2)
2
|e

|
2
|e

|
2
, (A1)
where r
e
= e
2
/m
e
c
2
, is the photon frequency, e and e

are the polarization vectors of the photon before and after

scattering. Equation (A1) retains only the resonance peak of the cross section and neglects the non-resonant part.
Positron cross section is described by the same equation except that e

is replaced by e
+
= 2
1/2
(e
x
+ie
y
) (positrons
gyrate in the opposite sense). The cross section can also be derived in the framework of quantum electrodynamics
(Herold 1979; Daugherty & Harding 1986). For
B
m
e
c
2
(which corresponds to B B
Q
) the result is reduced to
Equation (A1).
The resonance line is very narrow, /
B
= (4/3)(B/B
Q
) 1, where = e
2
/c = 1/137 (Daugherty & Ventura
1978; Herold et al. 1982), and the resonance factor [(
B
)
2
+(/2)
2
]
1
is well approximated by the delta-function
2
1
(
B
). Then the cross section may be written as
d
d

= 2
d
d

= 3
2
r
e
c (
B
) |e

|
2
|e

|
2
, (A2)
where +/ correspond to scattering by positron/electron,

= cos

, and

is the angle of the scattered photon with

respect to the magnetic eld B. The distribution of scattered photons is axially symmetric about B; this fact has
been used in Equation (A2).
Two polarization states (eigen modes) exist for the photon. They are controlled by the dielectric tensor of the
magnetosphere. In the considered region r < 100R, the dielectric tensor for photons of interest (X-rays) is dominated
by the magnetic vacuum polarization eect (e.g. Beresteskii et al. 1982); the plasma contribution to the dielectric
tensor is much smaller and may be neglected. Magnetic vacuum denes two linearly polarized eigen modes for
electromagnetic waves: e

which is perpendicular to the (k, B) plane, and e

which is parallel to the (k, B) plane

(here k is the photon wave vector and e shows the direction of the electric eld in the wave). The and modes are
also called E-mode and O-mode, respectively. Their refraction indices are
N

= 1 +
2
45
_
B
B
Q
_
2
sin
2
, N

= 1 +
7
90
_
B
B
Q
_
2
sin
2
, (A3)
15
where is the photon angle with respect to B. The two modes have slightly dierent propagation speeds c/N
and therefore they adiabatically track, i.e. the photon propagating through the curved magnetic eld preserves its
polarization state. The adiabaticity condition reads kl
B
(N

) 1 where l
B
r is the characteristic scale of the
spatial variation of B (see e.g. Fern andez & Davis 2011 for a detailed discussion). This condition is satised for X-rays
in the considered region of the magnetosphere where scattering occurs. Thus, in our transfer problem, the photon can
switch its polarization state only in a scattering event.
As the photon can be in either polarization state, calculation of radiative transfer involves four scattering processes
, , , and . The corresponding cross sections are given by Equation (A2) with |e

|
2
=
|e

|
2
= 1/2, |e

|
2
=
2
/2, and |e

|
2
=

2
/2. Note that the cross sections of electron and positron are
equal, as |e

|
2
= |e

e
+
|
2
for any linear polarization e.
Equation (A2) describes the cross section of electron (or positron) at rest. In our transfer problem, the particles are
moving along B and the above equations should be used in the rest frame of the particle. Note that the polarization
states and
B
are invariant under Lorentz boosts along B. Consider a photon with energy and propagation angle
with respect to B, in the or polarization state. We are interested in its scattering by an electron (or positron)
that moves with velocity c along the magnetic eld line. The total cross section in the lab frame may be obtained
by integrating the dierential cross section in the electron rest frame and then multiplying the result by 1 cos ,

tot
= 2
2
r
e
c (
B
) (1 ), (A4)
where
= (1 ) , (A5)
is the photon frequency in the electron rest frame and = cos . The factor depends on the photon polarization
and is given by

_
1,

2
,
_
, =

1
, (A6)
where = cos

and

is the photon angle with respect to B in the electron rest frame. The cross section
tot
is
summed over the nal polarization states.
The outcome of the scattering may be either or photon propagating at angle

in the electron frame. The

probability distribution for

= cos

is given by
P(

) =
1

tot
d
d

=
3
8

. (A7)
where

= 1 or
2
, depending on the nal polarization state. The integral
_
P(

) d

equals 3/4 for the state and

1/4 for the state. Thus, 3/4 of scattering events produce photons with uniform distribution P(

) = const and
the remaining 1/4 of scattering events produce photons with P(

)
2
. The distribution of the nal photons over
angle and polarization does not depend on the initial state of the photon before scattering.
The standard description of resonant scattering summarized above assumes the transition between the ground and
rst excited Landau levels, which has the largest cross section. Transitions to higher levels are neglected in our transfer
calculations.
A.2. Opacity
Optical depth of a relativistic plasma to resonant scattering was discussed previously in detail (e.g. Fern andez &
Thompson 2007 and refs. therein). Here we write down relevant equations and introduce notation that is used below
in the discussion of radiative drag. Consider a photon of energy that propagates through e

plasma with density

n = n
+
+ n

. The plasma particles move along B with momentum distribution f

e
(p) that is normalized to unity
_
f
e
(p) dp = 1. (A8)
The optical depth d seen by the photon as it propagates distance ds is given by
d
ds
= n
_

tot
f
e
(p) dp. (A9)
Assuming that the magnetosphere has a small optical depth to non-resonant scattering (Appendix C), we include only
resonant scattering and use Equation (A4) for
tot
. Integration over p may be carried out using the identity,
(
B
) =

k
(p p
k
)
|d /dp|
, (A10)
where
d
dp
=
d
dp
( p) = ( ), (A11)
16
and p
k
are all possible solutions of equation (p) =
B
. The delta-functions (p p
k
) express the fact that the
photon is scattered by electrons or positrons with momenta p
k
for which the resonance condition is met. The relation
sin

= sin together with the resonance condition =
B
determines sin

= (/
B
) sin and leaves two
possibilities for = cos

,
=
_
1

2

2
B
sin
2

_
1/2
. (A12)
Angle

exists (i.e. the resonance is in principle possible) for photons that satisfy the condition sin
B
. Then
Equation (A12) denes two electron velocities
1,2
, which may be found from the Doppler transformation of the photon
angle, = ( )/(1 ). It yields,
=

1
,
1,2
=
| |
1 | |
. (A13)
The corresponding Lorentz factors = (1
2
)
1/2
and dimensionless momenta p = are

1,2
=
1 | |
sin sin

, p
1,2
=
| |
sin sin

. (A14)
Substitution of Equations (A4) and (A10) to Equation (A9) and integration over p gives
d
ds
= 2
2
r
e
c

| |
n [f
e
(p
1
) + f
e
(p
2
)] . (A15)
Here = 1 for photons and =
2
for photons (Equation A6).
A.3. Radiative drag force
Consider now a radiation eld with intensity I(, k) in a given polarization state, or . As a result of scattering,
radiation exerts a force on the e

plasma. We are interested in the component of this force along the magnetic eld.
The force applied to unit volume of the plasma is given by
dP
dt dV
=
_
d
_
d
_
dp
I(, k)

nf
e
(p)
tot
P. (A16)
Here
_
d is the solid-angle integration over photon directions n = k/k, and P is the average momentum (per
scattering) passed to an electron or positron with Lorentz factor by a photon (, k). In the electron rest frame,

P equals the photon momentum along B, as its average momentum after scattering vanishes (resonant scattering
is symmetric in the electron frame when
B
m
e
c
2
). Thus,

P = /c and the Lorentz transformation of the
four-momentum vector to the lab frame gives
P =

B
c
, (A17)
where we used the condition =
B
since we consider only resonant scattering.
Radiation is described by two intensities I

(, k) and I

(, k) in the two polarization states. They scatter with

cross sections
tot
that dier by the factor of
2
(see Equations (A4) and (A6)). Substituting Equation (A4) to
Equation (A16) and taking the sum over polarizations, one nds the net force exerted by I

and I

,
dP
dt dV
=
_
d
_
d
_
dp
(I

+
2
I

nf
e
(p) 2
2
r
e
c (
B
)(1 ) P. (A18)
Integration over dp similar to that in Section A.2 gives
dP
dt dV
=
_
d
_
B sin
0
d 2
2
r
e

2
_
I

+
2
I

_
n[
1
f
e
(p
1
)
2
f
e
(p
2
)] , (A19)
where p
1,2
(, ) and
1,2
are given by Equations (A14) and (A12). The upper limit in the integral over takes into
account that only photons with
B
sin may be resonantly scattered (Section A.2). Equation (A19) is useful
because it shows the contribution of each photon (, k) (in the or polarization state) to the drag force. It can be
used even if the radiation intensity is not known in advance and needs to be found self-consistently with the plasma
dynamics (Section 6).
An alternative way of simplifying Equation (A18) is to rst integrate over , which gives
dP
dt dV
=
_
d
_
dp 2
2
r
e
(I

+
2
I

) nf
e
(p) ( ), (A20)
17
where I

and I

are evaluated at =
1
(1 )
1

B
. Equation (A20) is convenient to use when the radiation
intensity is known.
Note that Equations (A19) and (A20) are valid only where B B
Q
. Near the star, where the eld is stronger, the
drag force is modied (Baring et al. 2011; Beloborodov 2013). In this paper, we do not need the strong-eld corrections,
as radiative transfer occurs in the region of B B
Q
(where the scattered photons avoid the quick conversion to e

pairs); use of the full relativistic cross section would be an unnecessary complication.
B. OPTICALLY THIN OUTFLOW
Consider a magnetosphere that is optically thin to resonant scattering, so that intensity I is dominated by the
unscattered radiation from the star. We will assume that the neutron star emits approximately blackbody radiation
with the polarization (the photons dominate the surface radiation because they have a larger free path below the
surface, e.g. Silantev & Iakovlev 1980). Then
I

=

3
8
3
c
2
[exp(/kT) 1]
, I

= 0, (B1)
and we deal with the outow dynamics in the known radiation eld. This case was studied in detail in previous work
(e.g. Sturner 1995).
B.1. Scattering rate for one particle
Consider an electron (or positron) located at r, and moving outward with Lorentz factor = (1
2
)
1/2
along a
magnetic eld line. The number of photons scattered by the electron per unit time is

N
sc
=
_
d
_
d
I(, n)

tot
=
2
2
r
e
c

_
I(
res
[n], n)

res
d, (B2)
where we substituted
tot
(eq. A4). The resonant frequency depends on the photon direction,
res
=
1
(1n)
1

B
where n = k/k is the unit vector corresponding to d.
At large radii r R all photons at a given location r have approximately the same direction n r. The angle
between the stellar photons and the particle velocity, , is given by = cos B
r
/B (assuming > R/r). Then
integration over d in Equation (B2) is reduced to multiplication by (R/r)
2
, the solid angle subtended by the
star when viewed from radius r. This gives,

N
sc
=
2
3
r
e
c
x
2

I(
res
)

res

res
=

B
(1 )
, (B3)
where x = r/R. It is instructive to write

N
sc
in the following form,

N
sc
=

2
c
4x
2

g(y)
y
, (B4)
where = /m
e
c, = kT/m
e
c
2
10
3
and g(y) is the dimensionless Planck function evaluated at the frequency

res
=
1
(1 )
1

B
,
g(y) =
y
3
e
y
1
, y =

res
kT
=
b
(1 )
, (B5)
where b = B/B
Q
.
B.2. Drag force exerted on one particle
The drag force applied by the central blackbody radiation to the electron is F =

N
sc
P where P is given by
Equation (A17). This yields,
F() =

2
4x
2
m
e
c
2
r
e

3
g(y) (

), (B6)
where

= . Force F vanishes if =

; in this case the radiation ux measured in the rest frame of the particle
is perpendicular to B and cannot accelerate or decelerate it. In a weakly twisted magnetosphere, the magnetic eld
in the outer corona is approximately dipole. Then the saturation velocity

at a point r, (spherical coordinates)

depends only on and is given by

= =
B
r
B
=
2 cos
(1 + 3 cos
2
)
1/2
, p

=
2 cos
sin
. (B7)
The radiative force always pushes the particle toward p = p

. This eect may be measured by the drag coecient,

D
r
c
1
p
dp
dt
. (B8)
18
Consider an electron (or positron) with momentum p p

. A small deviation p p

causes drag D p p

,
which may be written as
D = D

_
1
p
p

_
, (B9)
where
D

=

2
4
R
r
e

3
g(y

)
x
2

4 10
4
x
g(y

_
kT
0.5 keV
_
3
. (B10)
Here y

= b

/ corresponds to photons that are resonantly scattered by the electron with p p

. The momentum
p

is a strong attractor if D

1. The value of D

is sensitive to y

. In particular, in the equatorial plane, we have

= 1 and
y

1.6 10
4
_
B
pole
10
15
G
__
kT
0.5 keV
_
1 _
r
R
_
3
, ( = /2), (B11)
where B
pole
is the dipole eld at the magnetic pole. The condition D

> 1 corresponds to y

<

20. This implies that

the e

ow on magnetic eld lines extending far from the star is stopped by the radiative drag in the equatorial plane.
For typical magnetar parameters, the ow stops on eld lines with R
max
>

10R.
B.3. Optical depth in the single-uid approximation
The single-uid ow has a distribution function f
e
(p

) = (p

p) where p(r, ) is the ow momentum. Then any

photon of energy may only be scattered on the innitesimally thin resonance surface dened by (1 ) =
B
,
where = cos describes the photon angle relative to the ow velocity. The optical depth of the resonant surface
may be obtained from Equation (A9), which gives
d
ds
= 2
2
r
e
c n (1 )(
B
), (B12)
where = (1 ). One can use the identity,
(
B
) =

k
(s s
k
)

d
ds
(
B
)

. (B13)
The location s
k
on the photon trajectory is where the photon crosses the resonant surface. Performing the integration
over s along the photon trajectory, one nds the optical depth for one crossing of the resonant surface,
=
2
2
r
e
c n (1 )

d
ds
(
B
)

. (B14)
If we specialize to the case of central photons emitted by the neutron star with the polarization,
6
then = 1 and
= B
r
/B. For a moderately twisted dipole magnetosphere, the electric current density is given by j c B/4R
max
(Beloborodov 2009), and
n = M
j
ev

MB
4eR
max
. (B15)
Note that n is small on eld lines with a large R
max
, which implies a low optical depth near the axis; this fact was
also emphasized by Thompson et al. (2002) who used a self-similar twist model.
The expression for the optical depth becomes particularly simple if the outow has the equilibrium momentum
p = p

. Then d /ds = 0, i.e. remains constant along the radial ray through the outow. This fact can be derived
by noting that the Doppler factor (1 ) =
1

is a function of only it does not depend on s = r for the

approximately dipole magnetosphere. It is also easy to see that d
B
/ds = 3
B
/r, and we nd
=

12
M
sin
4

cos (1 + 3 cos
2
)
1/2
. (B16)
The single-uid model with p = p

may approximate the outow only suciently close to the equatorial plane where
1

>

M
1
. Nevertheless, the approximate Equation (B16) shows a general feature: the optical depth seen by the
central photons is dramatically increased toward the equatorial plane ( sec ) and dramatically reduced toward the
axis ( sin
4
). As a result, a distant observer can see the unscattered radiation from the neutron-star surface when
the line of sight is within a moderate angle < /4 from the polar axis. This feature becomes even more pronounced
in the full radiative transfer problem where the relativistic outow is decelerated by the reected radiation from the
outer corona. Then scattering of the central radiation is negligible in the entire relativistic zone of the outow.
6
The neutron-star radiation is dominated by the polarization
(Silantev & Iakovlev 1980). In addition, for the outow with the
equilibrium momentum p considered below, the scattering of
photons (even if they were included) would be suppressed. In the
outow rest frame, the photons move perpendicular to B, and the
resonant cross section for the polarization mode vanishes.
19
B.4. Drag force exerted on a plasma with a broad distribution function
Equation (B6) describes the drag force exerted by the central thermal radiation on a particle with a given momentum
p = . One can also consider a collection of particles with a momentum distribution f
e
(p) and derive the average
force per particle

F = n
1
(dP/dV dt). Equation (A20) gives

F =
r
e

4c
2
_
R
r
_
2

3
B
_

( ) f
e
(p)

2
(1 )
3
(exp y 1)
dp. (B17)
where y is given in Equation (B5). The same result is obtained by averaging the force F given by Equation (B6),

F =
_
F(p)f
e
(p) dp.
Equation (B17) simplies when the plasma is described by the waterbag distribution function f
e
(Section 5.1); it
leads to a straighforward calculation of the ow dynamics in the central radiation eld. We use this simple outow
model as the rst trial to initiate the iterations that converge to the solution shown in Figure 9. In the nal solution,
the drag exerted by the central radiation turns out negligible in the relativistic zone; instead, the outow deceleration
is controlled by the radiation streaming from the equatorial reector, as discussed in Section 6. Then the force

F
derived in this section may be of interest only in the non-relativistic zone.
C. NON-RESONANT SCATTERING
Non-resonant scattering is not limited by the resonance condition, and hence many more photons can participate
in scattering, although with a smaller cross section. Below we discuss the eect of non-resonant scattering on the
dynamics of e

ow around magnetars.
Non-resonant scattering occurs mainly with photons in the Wien peak of the thermal radiation owing directly from
the neutron star, which dominates the photon density around the star. Relativistic particles see the thermal photons
(of typical energy E 3kT) blueshifted as

E = (1 )E where = cos describes the photon direction relative to
the particle velocity in the lab frame. Suciently far from the star (where R
2
/r
2
< 1 B
r
/B) the radiation can be
approximated as a narrow radial beam; then = B
r
/B. In general, is a function of the particle position r, , in
the magnetosphere. For an approximately dipole eld, = 2 cos (1 + 3 cos
2
)
1/2
is a function of the polar angle
only.
Magnetic eld strongly aects the non-resonant scattering cross section if

E <
B
= bm
e
c
2
. If the electron
is relativistic, the target photons are aberrated in the electron rest frame, cos

= = ( )/(1 ). In the
limit 1, even photons with the polarization have electric elds almost perpendicular to B, which makes their
scattering inecient. For photons with

E
B
, the non-resonant scattering cross section is given by (e.g. Canuto
et al. 1971)

_

E

B
_
2
+
sin
2

2
,

_

E

B
_
2
,

E
B
. (C1)
We assume

E m
e
c
2
and neglect Klein-Nishina corrections. Most of the radiation emitted by the neutron star has
the polarization.
The energy loss of the electron due to scattering is given by

E
e
=
_
d
_
dE (1 ) (

E)
I(n, E)
E
_
E

E
_
, (C2)
where n is the unit vector describing the photon direction in solid angle d, and E

=

E is the mean expectation for
the photon energy after scattering. This gives,

E
e
=
_
d
_
dE (1 )
_

2
(1 ) 1

I(n, E) (

E) dE. (C3)
In the simplest case of Thomson scattering of isotropic radiation, averaging over random gives the standard result

E
e
= (4/3)
T
c U
2

2
, where U is the energy density of radiation. In our case, <
T
, and the radiation eld is
not isotropic; far from the star it is better approximated as a central beam.
Using Equation (C3) one can show that non-resonant scattering makes a small contribution to the radiative drag
compared with resonant scattering, and hence its inclusion in the calculation does not signicantly change the outow
solution shown in Figure 9. Consider rst the non-relativistic zone p
+
< 1. An upper bound on the non-resonant

E
e
is obtained if we substitute into Equation (C3) (

E) =
T
and = 0. This gives,

E
e
=
T
c p
2
U. (C4)
The drag coecient due to non-resonant scattering is dened similar to Equation (B8). Using U L
th
/4r
2
c and
dp/dt =

E
e
/m
e
c
2
, one obtains
D

T
L
th

4 r m
e
c
3
0.2 L
th,35
r
1
6
1. (C5)
20
In the relativistic zone p
+
> 1, the upper bound given by Equation (C5) increases proportionally to and becomes
useless, because it does not take into account the strong reduction of the scattering cross section below
T
. In this
zone, the outow adjusts so that it can resonantly scatter photons with E 7kT and 0.5 (photons owing from
the equatorial reector). This implies that the main targets for non-resonant scattering (photons owing from the star
with E 3kT and > 0) have energies well below the resonance energy,

E (0.1 0.2)
B
. Then the scattering
cross section is strongly reduced below
T
according to Equation (C1). When this reduction is taken into account,
one obtains D < 1.
REFERENCES
Baring, M. G.; Wadiasingh, Z., & Gonthier, P. L. 2011, ApJ, 733,
61
Beloborodov, A. M. 1998, ApJ, 496, L105
Beloborodov, A. M. 2009, ApJ, 703, 1044
Beloborodov, A. M. 2011a, in High-Energy Emission from Pulsars
and their Systems, Astrophysics and Space Science
Proceedings, 299 (Springer-Verlag), arXiv:1008.4388
Beloborodov, A. M. 2011b, ApJ, 737, 68
Beloborodov, A. M. 2013, ApJ, 762, 13
Beloborodov, A. M., & Thompson, C. 2007, ApJ, 657, 967
Beresteskii, V. B., Lifshitz, E. M., & Pitaevskii, L. P. 1982,
Quantum Electrodynamics
Camilo, F., et al. 2006, Nature, 442, 892
Camilo, F., et al. 2007, ApJ, 666, L93
Canuto, V., Lodenquai, J., & Ruderman, M. 1971, Phys. Rev. D,
3, 2303
Cheng, A. F., & Ruderman, M. A. 1977, ApJ, 212, 800
Daugherty, J. K., & Harding, A. K. 1986, ApJ, 309, 362
Daugherty, J. K., &Harding, A. K. 1989, ApJ, 336, 861
Daugherty, J. K., & Ventura, J. 1978, Phys. Rev. D, 18, 1053
Fernandez, R., & Davis, S. W. 2011, ApJ, 730, 131
Fernandez, R., & Thompson, C. 2007, ApJ, 660, 615
Goldreich, P., & Julian, W. H. 1969, ApJ, 157, 869
Harding, A. K., & Daugherty, J. K. 1991, ApJ, 374, 687
Herold, H. 1979, Phys. Rev. D, 19, 2868
Herold, H., Ruder, H., & Wunner, G. 1982, A&A, 115, 90
Kardashev, N. S., Mitrofanov, I. G., & Novikov, I. D.1984, Sov.
Astronomy, 28, 651
Krall, N. A, & Trivelpiece, A. W. 1973, Principles of Plasma
Physics
Medin, Z., & Lai, D. 2010, MNRAS, 406, 1379
Lyubarskii, Y. E., & Petrova, S. A. 2000, A&A, 355, 406
Nobili, L., Turolla, R., & Zane, S. 2008, MNRAS, 386, 1527
Parfrey, K., Beloborodov, A. M, & Hui, L. 2012, 754, L12
Parfrey, K., Beloborodov, A. M, & Hui, L. 2013, ApJ, in press
(arXiv:1306.4335)
Rea, N., Zane, S., Turolla, R., Lyutikov, M., & Gotz, D. 2008,
ApJ, 686, 1245
Sturner, S. J. 1995, ApJ, 446, 292
Thompson, C. 2008a, ApJ, 688, 499
Thompson, C. 2008b, ApJ, 688, 1258
Thompson, C., & Beloborodov, A. M. 2005, ApJ, 634, 565
Thompson, C., Lyutikov, M., & Kulkarni, S. R. 2002, ApJ, 574,
332
Silantev, N. A., & Iakovlev, D. G. 1980, Ap&SS, 71, 45
Sturrock, P. A. 1971, ApJ, 164, 529
Ventura, J. 1979, Phys. Rev. D, 19, 1684
Woods, P. M., & Thompson, C. 2006, in Compact Stellar X-Ray
Sources, ed. W. H. G. Lewin & M. van der Klis (Cambridge:
Cambridge Univ. Press), 547