You are on page 1of 10

A&A 555, A101 (2013) Astronomy

DOI: 10.1051/0004-6361/201220966 &



c ESO 2013 Astrophysics

Modification of cosmic-ray energy spectra


by stochastic acceleration
R. C. Tautz1 , I. Lerche2 , and F. Kruse1

1
Zentrum fr Astronomie und Astrophysik, Technische Universitt Berlin, Hardenbergstrae 36, 10623 Berlin, Germany
e-mail: rct@gmx.eu
2
Institut fr Geowissenschaften, Naturwissenschaftliche Fakultt III, Martin-Luther-Universitt Halle, 06099 Halle, Germany
e-mail: lercheian@yahoo.com
Received 20 December 2012 / Accepted 22 May 2013

ABSTRACT

Context. Typical space plasmas contain spatially and temporally variable turbulent electromagnetic fields. Understanding the trans-
port of energetic particles and the acceleration mechanisms for charged particles is an important goal of todays astroparticle physics.
Aims. To understand the acceleration mechanisms at the particle source, subsequent eects have to be known. Therefore, the modifi-
cation of a particle energy distribution, due to stochastic acceleration, needs to be investigated.
Methods. The diusion in momentum space was investigated by using both a Monte-Carlo simulation code and by analytically solv-
ing the momentum-diusion equation. For simplicity, the turbulence was assumed to consist of one-dimensional Alfvn waves.
Results. Using both methods, it is shown that, on average, all particles with velocities comparable to the Alfvn speeds are acceler-
ated. This influences the energy distribution by significantly increasing the energy spectral index.
Conclusions. Because of electromagnetic turbulence, a particle energy spectrum measured at Earth can drastically deviate from its
initial spectrum. However, for particles with velocities significantly above the Alfvn speed, the eect becomes negligible.
Key words. plasmas magnetic fields turbulence acceleration of particles solar wind cosmic rays

1. Introduction Now the presence of electromagnetic turbulence gives rise


to reacceleration processes (Heinbach & Simon 1995) so that
Understanding the acceleration processes of cosmic rays has particles gain additional energy while propagating through
been a long-standing problem in astrophysics (e.g., Drury 1983; an otherwise homogeneous medium. The anomalous cosmic-
Blandford & Eichler 1987). It is currently accepted that high- ray component is also influenced by stochastic acceleration
energy particles with energies up to 1015 eV are accelerated processes (Moraal et al. 2008), thereby giving rise (Fisk &
in supernova remnants (e.g., Abdo et al. 2010; Hartquist & Gloeckler 2007a; Fisk et al. 2010) to a suprathermal tail in the
Morfill 1983) via diusive shock acceleration, thus producing distribution function with 5 as the energy spectral index. We
a power-law energy spectrum. The underlying processes date note, however, that thermodynamic arguments and the assump-
back to Fermi (1949) and have since been improved (Duy & tion of compressional turbulence may be sucient to derive a
Blundell 2005) by incorporating many additional eects (see similar spectral index (Fisk & Gloeckler 2007b). Likewise, adi-
Longair 2011, for an overview). The resulting energy spec- abatic cooling is a more ecient process than the interaction
trum agrees both with measurements (e.g., Nagano & Watson with turbulent electric fields (Zhang & Lee 2011; Litvinenko &
2000; Letessier-Selvon & Stanev 2011) and with laser experi- Schlickeiser 2011; Tautz et al. 2012). In anisotropic helical tur-
ments (e.g., Kuramitsu et al. 2011). Particles with low energies bulence, large-scale electric fields may also give rise to particle
up to 1010 eV, instead, originate from the Sun via coronal mass acceleration (Fedorov & Stehlik 2008).
ejections. For these reasons, much eort has been focused on the gen-
In either case, however, there are many processes that may eral properties of momentum diusion (e.g., Michaek et al.
change the energy spectrum of the particles on their way to 1999; Lee & Vlk 1975; Achterberg 1981) in slab and isotropic
Earth (Malkov et al. 2010). Energy-dependent loss and cooling turbulence with Alfvn and oblique propagating fast magne-
processes, for example, modify the energy distribution function tosonic waves, respectively. Furthermore, electrons near the Sun
(Buts et al. 2001), generally because the interstellar (and the in- may also be accelerated thanks to the interaction with Whistler
terplanetary) medium contains turbulent electric fields. Based waves (Hamilton & Petrosian 1992; Vocks et al. 2005, 2008).
on Mariner 2 measurements, it was found (Belcher & Davis Momentum diusion (Schlickeiser 2002, 1994) can be con-
1971) that, to some extent, the heliospheric plasma is domi- sidered as the underlying process of stochastic acceleration,
nated by outward-propagating Alfvnic turbulence, which has where particles mostly gain significant amounts of kinetic en-
been named the slab component. However, Alfvnic waves have ergy. If spatial diusion is to be neglected, stochastic accelera-
also been assumed for the investigation of cosmic-ray acceler- tion can be reduced to pure momentum diusion (Ostrowski &
ation at the galactic center (Fatuzzo & Melia 2011, 2012), for Siemieniec-Oziebo 1997). The connection of stochastic acceler-
example. ation and momentum diusion has been investigated analytically
Article published by EDP Sciences A101, page 1 of 10
A&A 555, A101 (2013)

using magnetohydrodynamic (MHD) waves (e.g., Skilling 1975; and (iii) the isotropic model, which does not assume a preferred
Schlickeiser 1989a) by applying quasi-linear theory (Shalchi & direction for the turbulent field. For reasons of simplicity, the
Schlickeiser 2004). slab model is used here.
In the presence of shock waves (Schlickeiser 1989b), the For the propagating plasma waves, which enter Eq. (3) via
first-order Fermi acceleration is usually dominant compared the wave frequency , specifically linearly polarized, undamped
to stochastic acceleration in the downstream region (Vainio slab Alfvn waves are chosen. In that case, the dispersion
& Schlickeiser 1999). In contrast, in this article the stochas- relation reads
tic acceleration outside a shock wave is investigated by means
of test-particle simulations. Such codes neglect the influence (k) = vA k , (2)
of the particles on the turbulence (Giacalone & Jokipii 1999; 
Michaek et al. 1999; Tautz 2010a), which is generally justi- where the Alfvn speed is vA = B0 / 4 with the mass
fied if concentration is focused on a tenuous component such as density. Other possible wave types such as slow and fast mag-
cosmic rays. Because the turbulence consists of MHD plasma netosonic waves or Whistler waves require three-dimensional
waves that carry an electric field, momentum diusion oc- turbulence, something that is beyond the scope of the present
curs so that the average particle energy is changed. In most investigation.
cases, particles with discrete energies are assumed so that the
energy-dependence of transport parameters such as the mean
free path can be investigated. Here, in contrast, an initial distri- 3. Monte-Carlo simulations
bution function is assumed, the time evolution of which is then For the test-particle simulations, the test-particle Monte-Carlo
traced so that the influence of the stochastic acceleration can code PADIAN (Tautz 2010a) is used, which solves the trajec-
be evaluated. Furthermore, the two cases will be distinguished tories of a large number of particles as they are being scattered
by (i) an ensemble that evolves without supplying new parti- by electromagnetic turbulence. Dimensionless variables are in-
cles; and (ii) continuous particle injection according to the pre- troduced as = t for the time with = qB/(mc) the Larmor
scribed initial distribution. Case (i) has been applied to solar en- frequency, and R = u/(0 ) as a dimensionless rigidity per
ergetic particle events (Drge 2005; Schlickeiser et al. 2009). magnetic field with 0 a characteristic length scale (see below).
Case (ii) corresponds to the generation of the anomalous cosmic-
ray spectrum (Fisk & Gloeckler 2006). Therefore, we intent
to demonstrate that Monte-Carlo test-particle simulations are a 3.1. Numerical turbulence generation
suitable tool for tracking the time evolution of a particle velocity Numerically, the turbulent electromagnetic fields are obtained
distribution or an energy-spectrum. from the superposition of a number of propagating MHD plasma
The paper is organized as follows: in Sect. 2, the basic prop- waves as
erties of the electromagnetic turbulence model are presented. In
Sects. 3 and 4, the numerical test-particle code PADIAN is intro- 
Nm
  
duced and the simulations results are presented, respectively. In B(r, t) = Re e A(kn ) exp i kn z (kn )t + n , (3)
Sect. 5, an analytical solution of the momentum-diusion equa- n=1
tion is derived, and is compared to the numerical results. A brief
where the wavenumbers kn are distributed logarithmically in the
summary and a discussion of the results with regard to future
interval kmin  kn  kmax and where is a random phase angle.
applications are given in Sect. 6.
For
the amplitude and the polarization vector, one has A(kn )
G(kn ) and e ez = 0, respectively, with the primed coordinates
2. Astrophysical turbulence properties determined by a rotation matrix with random angles.
With these assumptions and for the variables introduced
In astrophysical plasmas such as the heliosphere, turbulent elec- above, the Newton-Lorentz equation can be expressed as
tromagnetic fields are generated by instabilities or by large-

scale motion cascaded towards smaller scales. Measurements d B B
in the solar wind (e.g., Bruno & Carbone 2005) confirm the R = RA e E + R e B0 + e B , (4)
d B0 B0
Kolmogorov (1991) power law, which gave rise to a spectral tur-
bulent energy distribution of the form (e.g., Bieber et al. 1994; where e B and e B0 are unit vectors in the directions of the tur-
Tautz et al. 2006b) bulent and the background magnetic fields, respectively. The
 5/6 unit vector for the electric field e E is obtained using Faradays
G(k) 0 (B)2 1 + (0 k)2 , (1) induction law (Schlickeiser 2002). Furthermore, the parameter
RA vA /(0 ) is called the Alfvn rigidity.
where 0 = 0.03 a.u. denotes the turbulence bend-over scale.
In addition to the turbulence spectrum, the particle behavior
is also influenced by the turbulence geometry, i.e., the angular 3.2. Diffusion coefficients
distribution with respect to a preferred direction, which is given In the normalized coordinates, spatial diusion coecients are
by an ambient magnetic field such as the solar magnetic field defined through the mean square displacements in the direc-
(Parker 1958; Fisk 1996). Apart from fully three-dimensional tions parallel and perpendicular to the ambient magnetic field
anisotropic models (Matthaeus et al. 1990; Weinhorst & Shalchi as (Tautz 2010a)
2010; Rausch & Tautz 2012), there are three models that have
been used over the years: (i) the slab model, which assumes , (t) 1  2
= x , (t) . (5)
that the turbulent field depends solely on the coordinate along 02 2t
the mean magnetic field B0 as B = B(z) for B0 = B0 e z ;
(ii) the composite model, which superposes the slab field by a Here, is time-dependent and hence is frequently denoted the
two-dimensional complement as B = Bslab (z) + B2D (x, y); running diusion coecient.
A101, page 2 of 10
R. C. Tautz et al.: Modification of cosmic-ray energy spectra by stochastic acceleration

Using the same general derivation, a dimensionless run- Table 1. Parameters used for the Monte-Carlo simulation.
ning momentum diusion coecient can be defined through (cf.
Ostrowski & Michaek 1993; Tautz 2010b) Parameter description Symbol Value
Dp (p, t) 1 Minimum rigidity Rmin 103
= [R(t) Rinit ] .
2
(6) Maximum rigidity Rmax 1
p2init R2init t Alfvn rigidity RA 104; 3
Energy spectral index a 1.5
From the inverse momentum diusion coecient, the so-called Number of plane waves N 256
acceleration time scale (e.g., Zank et al. 2006; OSullivan et al. Number of particles 105
2009; Dosch & Shalchi 2010) can be constructed as Relative turbulence strength B/B0 1

p p2 Notes. The number of particles used for the averaging process is


acc (7) subdivided into 250 dierent turbulence realizations corresponding to
p Dp (p)
dierent sets of random numbers, in each of which the trajectories of
Whether or not the momentum diusion coecient is truly dif- 400 particles are calculated.
fusive, i.e., whether it has a finite constant value for large times,
is not entirely clear at present (cf. Tautz 2010b; Tautz & Shalchi
2010). Generally, the larger the ratio RA /R, the sooner Dp attains
an asymptotic value1 .

3.3. Rigidity distribution


In most test-particle simulations, a number of discrete initial
particle rigidities are assumed which are then investigated sepa-
rately, which is acceptable when one attempts to investigate the
rigidity-dependence of the diusion coecients. Here, however,
the goal is to trace the evolution of a given rigidity distribution
function, while the particles are scattered in momentum space.
In accordance with cosmic-ray observations (see
Letessier-Selvon & Stanev 2011, for recent observations),
the initial rigidity distribution is modeled in the form of a
power law. The requirement that the distribution function be
normalized to unity leads to
1a
f (R, a) = Ra (8a)
R1a
max R 1a
min

in Rmin  R  Rmax and f (R, a) = 0 otherwise. We note that


Eq. (8a) is valid for a  1, whereas for a = 1 one has the special
form

1
Rmin 1
f (R, 1) = ln (8b)
Rmax R
again in Rmin  R  Rmax . For the numerical simulations, the
spectral index has to be chosen so that, on the one hand, a rel-
atively steep initial energy spectrum is obtained. On the other Fig. 1. Evolution of the particle distribution as a function of the rigid-
hand, the steeper the spectrum, the fewer particles will be found ity R as the normalized time t increases. For the Alfvn rigidity, val-
at the high-energy end of the spectrum, thus requiring a larger to- ues of RA = 104 (upper panel) and RA = 103 (lower panel) have been
tal number of particles. As a compromise, here the value a = 1.5 used. (This figure is available in color in electronic form.)
will be used. The same holds true for the minimum and maxi-
mum rigidity values. Whereas the interpretation of the numeri-
cal simulations requires a large number of particles per rigidity method (e.g., Press et al. 2007, p. 362). The method requires uni-
interval, Rmin and Rmin are chosen so that particles with rigidities form random numbers x [0, 1] to be fed to the inverse cumulant
comparable to and much larger than RA are present but, at the F of the distribution function, which reads
same time, the rigidity interval is kept moderately narrow.
In the simulation code, particle rigidities are randomly drawn   
1a 1/(1a)
F 1 (x, a) = R1a
min + x Rmax Rmin
1a
(9a)
from the distribution function (Eq. (8a)) using the transformation
1
At first glance is might seem puzzling that diusion coecients are for a  1 and
not constant even though there is turbulence from the start. A long time
ago, when comparing random particle motions with diusive processes, F 1 (x, 1) = Rmax
x
R1x
min (9b)
Chandrasekhar (1943) found that truly diusion-like behavior the corre-
sponding equivalent diusion coecient is time-dependent and reaches if a = 1, thereby returning rigidity values R with the probability
an asymptotic constant value only later in time. prescribed by Eq. (8a).
A101, page 3 of 10
A&A 555, A101 (2013)
4
10
0.016

0.014
R3.26
0.012

0.01

Rpeak
3
10
f(R)R

0.008
R2.77
0.006

0.004

1.50
R 0.002
2
10 0
0 2000 4000 6000 8000 10000
t
0.1
3 2 1 0
10 10 10 10
R 0.08

4 0.06
10

Rpeak
R4.96 0.04

R1.50
3
10 0.02
f(R)R

0
0 2000 4000 6000 8000 10000
2
R4.35 t
10
Fig. 3. Maximum of the distribution function as the time increases.
Again the cases of RA = 104 (upper panel) and RA = 103 (lower
panel) are shown. The black and red lines denote the cases of single
1
injection and continuous particle injection, respectively. The thick lines
10 simply show cubic fits of the actual simulation data. Additionally, the
3 2 1 0
10 10 10 10
R average particle rigidity is shown in the lower panel (dotted line). (This
figure is available in color in electronic form.)
Fig. 2. Energy spectral index for the cases of RA = 104 (upper panel)
and RA = 103 (lower panel). Shown is the energy distribution function
at the beginning (black lines) and at the end of the simulation for the
cases of single injection (red lines) and continuous particle injection spectra. Whereas initially all particles are distributed according
(blue lines). Also shown are the spectral indices obtained from power- to the power law in Eq. (8a), this is later changed towards a bro-
law fits (dashed lines) to the simulation data (thin lines). (This figure is ken power-law distribution (upper panel) and, for larger Alfvn
available in color in electronic form.) velocities, a distribution that shows a power-law feature only in
a very narrow energy range. In all simulation runs, the spectral
indices have been fitted to the steepest range. The notable re-
4. Simulation results sult is that in the case of continuous injection the spectrum is
only moderately flattened even though newly injected particles
Two basic sets of simulations need to be distinguished: (i) sin- are present at all times.
gle injection, where all particles are injected at time t = 0 and The maximum particle rigidity is depicted in Fig. 3, show-
their trajectories are monitored; and (ii) continuous injection, ing a steady increase due to stochastic acceleration. The only
where, throughout the simulation, new particles are injected with exception is found for low Alfvn rigidity in the case of contin-
rigidities determined according to the distribution function from uous injection, where the average rigidity remains constant for
Eq. (8a). Technically; case (ii) is realized by choosing the max- some time. Again, it is the Alfvn rigidity that has the prominent
imum simulation time as a random number so that, throughout influence on the resulting average particle energy, at least qual-
the simulation, particles with random life times are present. All itatively, and not the average particle life-time as distinguished
simulation parameters are summarized in Table 1. by continuous versus single injection.
In Fig. 1, the evolution of the distribution function is traced The associated running momentum diusion coecient de-
as time increases. Initially, a power-law distribution according to termined through Eq. (6) is shown in Fig. 4. We note that the
Eq. (8a) is assumed. As the simulation time increases, a maxi- time is normalized as vA t/0 , which illustrates that for smaller
mum in the distribution function is quickly established. Because Alfvn velocities larger times are needed until the momentum
of the stochastic acceleration of the particles, the maximum is diusion coecient reaches its asymptotic limit. It is precisely
then constantly shifted towards higher energies. In the case of a the fact that almost a 1/t time dependence can be seen before
low Alfvn velocity, the process proceeds relatively slowly. We finally a transition to a constant (diusive) value occurs that
note that the results shown in Fig. 1 correspond to case (i). makes it so numerically dicult to obtain a reliable value for
The quantitative variation of the energy spectral index is il- the momentum diusion coecient. The maximum simulation
lustrated in Fig. 2, which illustrates the time evolution of the ve- time vA t/0 = 102 corresponds to t = 105 and t = 106 , re-
locity distribution by comparing the initial and the final particle spectively, which illustrates that for the parameters chosen here,
A101, page 4 of 10
R. C. Tautz et al.: Modification of cosmic-ray energy spectra by stochastic acceleration
2
10 here for reasons of simplicity. We note that, by defining a di-
RA = 103 mensionless rigidity diusion coecient DR = Dp /(02 m2 3 ),
RA = 104 Eq. (10) can be written in dimensionless form by using the
3 normalized variables introduced in Sect. 3.
10
For Alfvn waves, the diusion coecient (see, e.g.,
Schlickeiser 1985, 1989a; Stawarz & Petrosian 2008) reads
Dp

2  2 2
1s
p2
10
4 B vA p c 2
Dp (p) (11)
B0 c R2sL
kmax

5
with RL the Larmor radius, so that Dp p s . It should be noted
10 however that Eq. (11) was derived only for a simple turbulence
spectrum of the form G(k) ks with s [1, 2] (for a more
detailed discussion see, e.g., Schlickeiser 2002, Sect. 13.1.3).
10
6 Consider the diusion coecient written as Dp = D0 pq and
10
1
10
0
10
1 2
10 the initial distribution written in the form f (p, a) = f0 pa . Then
vA t/0 Eq. (10) is given as


Fig. 4. Numerically calculated running momentum diusion coecient f D0 q+2 f
for the cases of RA = 104 (solid line) and RA = 103 (dashed line). We = 2 p (12)
note that for the last part of the lower curve, only 103 particles could be t p p p
used, while the rest was obtained from a simulation with 104 particles.
The two diamonds illustrate the quasi-linear result from Eq. (B.2) in
In general, there are two options. First, when f is initially pre-
Appendix B. scribed as the power law in Eq. (8a), then a Fourier approach
can be used as shown in Appendix A. This allows the reduction
of the momentum diusion equation to an ordinary dierential
scattering in momentum takes an extremely long time until the equation but has the disadvantage that the Fourier transform of
diusive limit is reached2. the initial distribution function must be known. Second, when f
is initially unknown, a self-similar solution can
 be sought,
 which
5. Analytical stochastic acceleration is based on the separation f (t, R) = S (t)H p/E(t) and will be
discussed in the next subsection.
In principle, the stochastic variations in the particle energy may
be positive or negative. However, as shown in Sect. 4, the net
eect was a significant energy gain of the particle ensemble. To 5.1. Self-similar solution
verify this finding, we present an approach to this problem by Consider again Eq. (12). For a self-similar solution one seeks a
directly solving the equation of momentum diusion. The be- solution in the form
havior of charged particles in turbulent electromagnetic fields

has to be described using a Fokker-Planck approach (Parker p
f (t, p) = S (t)H , (13)
1965; Schlickeiser 2002) or at least a diusion-convection equa- E(t)
tion. Usually, the terms that describe spatial diusion are of
higher magnitude than momentum-diusion terms. For partic- where S , H, and E need to be determined. With = p/E(t),
ular parameter values analytical solutions are feasible (e.g., Eq. (12) can then be written
Bsching et al. 2005; Pohl 2009), whereas a full solution 

requires numerical tools (Guyer et al. 2009). 1 f  f E D0 q+2 f
 + = 2 (14)
However, if concentration is to be focused on stochastic en- E 2q t  E 3q
ergy changes only (e.g., Sect. 3.3 in Tverskoi 1967; Blandford
& Eichler 1987; Ostrowski & Siemieniec-Oziebo 1997), it is which, by using Eq. (13), yields
sometimes convenient to neglect spatial diusion altogether.

Under certain conditions3 the diusion-convection equation can S E dH D0 S d q+2 dH
H 2q S 3q = 2
then be expressed as E E d d d


f 1 f
= 2 p2 Dp (p) , (10) 5.1.1. Solution for general q  2
t p p p
which is a linear parabolic partial dierential equation. In princi- Collecting terms, one has
ple, Dp depends also on time, which however will be neglected

S E dH D0 d q+2 dH
= 2 , (15)
2
Of course, long is a relative term as 106 gyro periods is a short time E 2q S E 3q H d H d d
compared to the dynamical time scales in most astronomical sources.
However, both computationally and with respect to spatial diusion which is separable when (the coecients C1 , C2 , and C3 are
which takes about a factor of 104 fewer gyro periods that time scale is constants)
indeed extremely long.
3
For example, the principle of detailed balance has to be fulfilled, E
which requires an equilibrium state (e.g., Klein 1955). Specifically, = C1 (16a)
E 3q
in the present case it states that the probability (p, p ) for changing
the particle momentum from p to p satisfies the condition (p, p ) = dE q2
= C1 (2 q) (16b)
(p , p). dt

A101, page 5 of 10
A&A 555, A101 (2013)

so that
4
10
 
E = C2 C1 (2 q) t 1/(2q) . (16c)
2
Then additionally one must have 10
t=0
S S  
= C2 C1 (2 q) t C3 , (16d)

f(p)
0
E 2q S S 10
t=1
which integrates trivially so that one obtains S (t) in the form
  2
10
S (t) = S 0 C2 C1 (2 q) t C3 /[C2 (2q)] (17) t=2
and one has 4
10


d q+2 dH C1 3 dH C3 2
+ H = 0. (18) 4 3 2 1 0 1 2
d d D0 d D0 10 10 10 10 10 10 10
p
For further simplification use the fact that the distribution func-
Fig. 5. Evolution of the momentum distribution function as obtained
tion is normalized, i.e.,
from the self-similar solution of the momentum diusion equation for
 the case q = 2. The three lines illustrate the distribution at the times t =
1 = S (t)E(t) 3
d 2 H() S (t)E(t)3 H0 (19) 0 (black solid line), t = 1 (blue dashed line), and t = 2 (red dotdashed
0 line). (This figure is available in color in electronic form.)
so that S (t)1 = E(t)3 H0 , yielding S /S = 3E/E
and

S E The solution for the distribution function is then


= = C3 = 3C1 .  
E 2q S E 3q
f (t, p) = S 0 E0a exp D0 a2 t exp (3atD0 ) pa (24a)
Finally, the resulting dierential equation for H(t) reads
= S 0 E0a pa exp [D0 a (3 a) t] (24b)


d q+2 dH C1 2 d(H)
+ = 0, (20) in min exp(D0 at)  p  max exp(D0 at). From the values at t = 0
d d D0 d
one therefore has min,max = pmin,max (t = 0).
which needs a numerical solution except for special values of q From the normalization condition (Eq. (23c)) one can choose
such as q = 2. E0 = 1 so that
3a
5.1.2. Solution for q = 2 S0 = (25)
(max (min )3a
)3a
For instance when q = 2, set H = so that Then the distribution function f (t, p) has the same normaliza-

tion for all times (as must be the case); however, the minimum
C1 C3
+ 3+
2
= 0, (21) and maximum momenta increase as time increases. In Fig. 5, the
D0 D0 distribution function is shown for dierent times, illustrating the
yielding shift of the particles toward higher momenta.
 We note that Eq. (12) shows that for q = 2 there is scale inde-


2 pendence of the momentum (as can easily be seen by replacing
C1 C1 4C3
2 = 3 + 3+ + (22) p by Lp where L is an arbitrary constant). Equation (12) is un-
D0 D0 D0 altered by such a replacement. Thus if the initial distribution has
no scale, as with a power law, then the distribution never devel-
Now at t = 0, the distribution function f is discontinuous; there- ops a scale and retains the power law. If, however, the initial dis-
fore, let H = a in min   max and H = 0 otherwise, tribution is terminated at specific values of momenta then these
where min and max are fixed values to be related eventually end values must adjust with time to maintain the scale indepen-
to pmin and pmax . dence and also to maintain a constant particle population. This
The normalization requires anomalous result happens only for q = 2; for any other value of q
 
there is an overall scale for the momentum in Eq. (12) so that the
p
1= dp p f (p) = S
2 2
dp p H (23a) particle evolution with time will end up not only with adjusted
0 0 E(t) end point values but also with a change in the spectral index, as
 max
indeed is shown by the numerical illustrations. Furthermore, we
= S (t)E(t)3 d 2 H() (23b) note that the case q = 2 agrees with the so-called hard-sphere
min
scattering (e.g., Parker & Tidman 1958; Park & Petrosian 1995)
S (t)E(t)  3 
and thus possesses a tangible physical interpretation.
= (max )3a (min )3a . (23c)
3a Finally, the average momentum gain per time can be
obtained using
Hence, S (t)E(t)3 must be constant, which again relates the co-
ecients C3 = 3C1 . Then from Eq. (21) with = a one d
has C1 = D0 a. ln p(t) = D0 a, (26)
dt
A101, page 6 of 10
R. C. Tautz et al.: Modification of cosmic-ray energy spectra by stochastic acceleration
1
10 The solution allows one to predict the time evolution of the
distribution function. A general method to obtain a solution
by Fourier transforming the momentum-diusion equation is
0
presented in Appendix A.
10
p p0 

By monitoring the average momentum value, it was confirmed in


p0 

both cases that particles are eciently accelerated. Furthermore,


the simulation results show that, for a distribution function that
is initially of the form f pa , the spectral index a is signif-
1
10
icantly increased for later times. The steepened spectrum is in
remarkable agreement with p5 spectra observed in various sce-
2 narios such as pick-up ion distributions (Fahr & Fichtner 2011),
10
termination-shock particles (Jokipii & Lee 2010), and the so-
lar corona (quiet solar wind particles; Fisk & Gloeckler 2008;
Antecki et al. 2013). We note that, because steep spectra take
10
3 considerably more computing time, a relatively flat initial spec-
0 0.2 0.4 0.6 0.8 1 trum has been chosen; however, the agreement with observa-
t tions is of itself no guarantee that this agreement cannot also be
Fig. 6. Relative average momentum gain as described through Eq. (27) achieved with a steep injected spectrum. Demonstrating the va-
as a function of time. The reference value is the initial average momen- lidity of this point is, however, best delayed to a future communi-
tum, i.e., p0 = p(t = 0) . cation. In addition, here the acceleration is due to homogeneous
turbulent electromagnetic fields generated by (unstable) plasma
waves. This steepening of the spectrum could not be reproduced
where in the analytical self-similar solution based on a momentum dif-

fusion coecient of the form Dp pq with q = 2 for reasons
p(t) = dp p3 f (t, p) (27a) given in the text. This spectral index change requires a depar-
0
ture from a scale-independent momentum behavior for Eq. (12),
(pmax )D0 a (pmin )4 (pmax )4 (pmin )D0 a something that occurs for all values of q except q = 2. Therefore,
=
(pmax )D0 a (pmin )3 (pmax )3 (pmin )D0 a the latter case has been chosen for reasons of simplicity.
3 D0 a D0 at Furthermore, the results depend on the details of the
e . (27b) momentum-diusion coecient Dp for which a simplified form
4 D0 a
is assumed. As detailed in Appendix B, a more general, non-
The relative momentum change is shown in Fig. 6, thereby il- linear approach shows that Dp exhibits a more complicated de-
lustrating that particles are eciently accelerated in agreement pendence on the Alfvn velocity and on the turbulence power
with the simulation results shown in Sect. 4. Therefore, it has spectrum. In addition, the cases of single and continuous parti-
been confirmed that, on average, particles indeed gain energy, cle injection have been considered in the simulation. Whereas
thus confirming the numerical ansatz. the second case might be more realistic, the first case is useful
in that, by superposition of dierent time injection solutions, a
6. Discussion and conclusion continuous injection model can be constructed where the source
term varies with time. This, however, is beyond the scope of the
In this paper, the stochastic acceleration of energetic charged present investigation.
particles, due to momentum diusion in Alfvnic electromag- Future work should attempt to modify the analytical self-
netic turbulence, has been investigated. The initial particle dis- similar solution so that the eect of a variable energy spectral
tribution function was prescribed as a power law starting with index is incorporated as predicted by the numerical simulations.
the Alfvn velocity and covering three orders of magnitude. Generally, a much larger number of test particles than is cur-
Even though it was found (Schlickeiser 1989a) that the domi- rently available is required if the evolution of a distribution func-
nant contribution to the stochastic acceleration of charged parti- tion is to be investigated for a broader momentum range and/or
cles is provided by fast magnetosonic waves and not by Alfvn for a considerably steeper initial spectrum. This is especially true
waves, here the case of Alfvn waves was chosen because these when the particles are continuously injected so that the evolu-
waves constitute the dominant time-dependent turbulent compo- tion of a quasi-equilibrium can be monitored. Furthermore, real
nent in the solar wind. Throughout, radiative losses (Schlickeiser physical scenarios such as particle acceleration at the outer re-
& Lerche 2007; Stawarz & Petrosian 2008; Schaefer-Rols et al. gions of the solar system require the inclusion of spatial struc-
2009) have been neglected as only wave-particle interactions tures, thereby giving rise to spatially variable diusion coe-
were included via the Lorentz force (cf. Tautz 2010a). cients. The investigation of this problem will be deferred to a
Two independent methods were used: future paper.
A Monte-Carlo simulation, which, in the presence of ar- Acknowledgements. F.K. thanks D. Breitschwerdt for support during his M. Sc.
tificial turbulence obtained from the superposition of slab thesis. We are also appreciative of comments by a referee that helped us sharpen
some otherwise fuzzy points.
Alfvn waves, integrates the Newton-Lorentz equation for
a large number of test particles. Additionally, the cases were
distinguished of all particles being injected (i) at the initial Appendix A: Fourier solution
time t = 0; or (ii) continuously throughout the simulation of the momentum-diffusion equation
with probabilities given by the initial distribution function.
By neglecting spatial diusion, the momentum-diusion The analytical solution of Eq. (10) for the initial distri-
equation has been analyzed using a self-similar approach. bution from Eq. (8a) proceeds as follows. There are two
A101, page 7 of 10
A&A 555, A101 (2013)
9
alternatives: (i) a direct solution via Fourier transformation; and 10
(ii) a self-similar solution approach. While the second approach
is discussed in the text, here some basic considerations are Dp
shown regarding the first approach. p2
Set d = p(q+2) dp so that = p(q+1) /(q + 1) when Eq. (12)
takes the form 10
10 R 0.4s
f D0 2 f
= q+4 2 (A.1)
t p
With the Fourier transform dependence in time proportional
to eit , Eq. (A.1) yields 10
11
3 2 1 0 1
10 10 10 10 10
i q+4 2 F R
p F= 2, (A.2)
D0 Fig. B.1. Quasi-linear result for the momentum diusion coecient
where F(, ) is the Fourier transform of f . Now (solid line). The dotted line is the power-law fit, which yields Dp p0.4s .
 (q+4)/(q+1)
1
pq+4 = which, at t = 0, has a dependence as f (0, ) (2+1)(m2) . Now
(q + 1)
originally, f (0, p) is given in the form f pa and p(q+1) .
so that Eq. (A.2) can be written From Eq. (A.8) it then follows that

(q+4)/(q+1)
i 1 2 F 1 a
(q+4)/(q+1) F = 2 = + (A.9)
D0 q + 1 2 2 (q + 1) (m 2)

With = i/D0 [(q + 1)](q+4)/(q+1) and m = (q + 4)/(q + 1) which connects the expression for f with the frequency power-
one has law F0 () . And so one has the solution, provided no lim-
its are set on the pa behavior. Furthermore, for finite times,
2 F t > 0, one then has a relatively simple integral for f (t, ) from
+ m F = 0, (A.3) 
2 Eq. (A.8). We note also that, if F0 () = fn n then superpo-
sition of inverse Fourier integrals of the above type gives the
which has the solution

general solution. Thus for any f (p) that is expandible in or-
 2 (2m)/2 thonormal functions expressed as power combinations through a
F(, ) = F0 () J1/(2m) , (A.4) Gram-Schmidt orthonormalization process one has the solution
2m
by superposition.
where Jn (z) is the Bessel function of the first kind of order n and
where F0 () is constant in . Then


 
Appendix B: Non-linear momentum diffusion
2 (2m)/2
f (t, ) = d F0 ()eit J1/(2m) . (A.5) With RA /R, the general expression for the Fokker-Planck
2m
coecient of momentum diusion is of the form
We note that 2 m = (q 2)/(q + 1). 
Suppose that F0 () is a simple power in , i.e., F0 () = 2 p2 2 1  
Dp = d 1 2
dk G(k )
f0 (i) . Now set = 0 (i) so that 2B20 1
 
 
  
(v jvA ) k + + (v jvA ) k , (B.1)
f (t, ) = f0 d (i) e(i)t
j=1


2 0
J1/(2m) 2m (i) . (A.6) which was derived under the assumption of slab linearly polar-
2m ized, undamped Alfvn waves (Schlickeiser 2002).
Using quasi-linear theory (Jokipii 1966), the momentum dif-
Set i2m 2 so that 2 d = id 2m and = fusion coecient for Alfvn turbulence with vanishing cross
(2m)/2 e3i/4 1/2 . Careful inspection of the integration limits helicity is given through (Schlickeiser 2002)
shows that

2  1  1 2
1
(2m)/2 ei/4 E1 (A.7a) Dp 2 B 1
= C(s) d
G , (B.2)
p2 R B0 0 | j| R | j|
+ j=1
(2m)/2 e3i/4 E2 . (A.7b)
where
The resulting inverse Fourier integral of the distribution function 1 (s/2)
then reads C(s) =  (B.3a)
2 (1 s)/2)
 E2
f (t, ) = 2i f0 (2+1)(m2)
d 2+1 ensures the normalization of the spectrum from Eq. (1), from
E1 which only the wavenumber-dependent part

 
2 0  s/2
J1/(2m) exp 2 m2 t , (A.8) = 1 + (0 k)2
2m G(k) (B.3b)
A101, page 8 of 10
R. C. Tautz et al.: Modification of cosmic-ray energy spectra by stochastic acceleration

is used here. For the spectral index, s = 5/3 is chosen in accor-


dance with Eq. (1). In the limit R  1, the resulting momentum
diusion coecient can be fitted by a power law Dp p0.4s (see
Fig. B.1).
To obtain an average momentum diusion coecient for a
distribution function from Eq. (8a), an expectation value has to
be calculated as
 Rmax
Dp = dR Dp (R) f (r, A). (B.4)
Rmin

For the parameters given in Table 1, one has


2.2 106 , for RA = 104
Dp = (B.5)
3.3 104 , for RA = 103 ,
which is in approximate agreement with the simulation results
shown in Fig. 4. Fig. B.2. Comparison of the momentum diusion coecient Dp as ob-
A recent investigation of Shalchi (2006) using the weakly tained from numerical test-particle simulations (triangles), quasi-linear
theory (dashed line), and second-order quasi-linear theory (crosses).
non-linear theory (Shalchi et al. 2004) showed that, compared
to previous, quasi-linear results (Shalchi & Schlickeiser 2004),
the velocity-dependence of the momentum diusion coecients is the second-order resonance function. The integrals in
is flattened. Here, the second-order quasi-linear theory (SOQLT, Eq. (B.8a) cannot be solved analytically so that a numerical so-
see Shalchi 2005; Tautz et al. 2008; Tautz & Lerche 2010) is lution, which in this case is intricate, has to be evoked.
The result for Dp is shown in Fig. B.2 in comparison to re-
used to derive the non-linear equivalent to judge the resulting
dierences. sults obtained from a test-particle simulation and the quasi-linear
result from Eq. (B.2). At first view, the SOQLT result seems
While in QLT, particles are assumed to follow the unper-
to be an improvement compared to QLT for rigidities larger
turbed spiral trajectory, SOQLT takes into account the distribu-
than R = 0.0015 but one has to be careful before coming to
tion of particles around their unperturbed orbits through reso-
any conclusions. A conspicuous feature of the values obtained
nance broadening. For the Fourier transform, the new approach
with SOQLT can be seen at the lower rigidity range. The val-
involves deviations from the unperturbed orbit, x (t), as
ues exhibit a discrepancy in comparison to the simulation, be-
ik x (t)
cause the slope of the curve is positive for rigidities smaller than


e

, QLT
R = 0.0015. As was pointed out in the preceding, the simula-
e ikx
=
(B.6)


dz f (z)eikz , SOQLT, tion in this parameter regime provides the most reliable results,
since for small rigidities it reached almost diusive behavior.
where a Gaussian shape is assumed for the distribution function Therefore, one would expect a decreasing slope everywhere and
so that the values for the Fokker-Planck coecient to be considerably
  higher in this range. In the calculation, the Alfvn rigidity is set
1 (z z(t) )2 to RA = 103 in the calculation, hence the approximation is no
f (z, t) = exp (B.7) longer valid in this regime.
2(t) 22z (t)
There is an additional feature of the simulations that has to
with z(t) = z(t) = 0 the (vanishing) mean of the unperturbed be taken into account. As shown in Fig. 4, the momentum diu-
parallel coordinate. sion coecient decreases for a longer time if the ratio = RA /R
The mean square parallel coordinate z2 (t) that enters the is decreased, i.e., for larger initial particle rigidities. Therefore,
width of the Gaussian as 2z (t) = z2 (t) z(t) 2 is obtained using longer simulation times would be required in principle to en-
QLT, thereby justifying the designation of the resulting theory as sure that the diusive regime is reached for all initial rigidities.
being of the second order. In the limit of large times t  1 and Therefore, the values of the momentum diusion coecient are
for pitch-angles close to 90 , simple analytical expressions can overestimated for larger initial rigidities. However, quantitative
be obtained for z2 (t) , thereby modifying the resonance func- statements about the expected reduction are not possible; in-
tion. This has a deep impact on some transport parameters. For stead, simulations with longer running times would be required,
instance, the mean free path in isotropic magnetostatic turbu- which are beyond what is currently available.
lence (Tautz et al. 2006a, 2008) is infinitely large in QLT, but in
SOQLT agrees well with simulations (Tautz 2012).
For the momentum-diusion coecient, the result reads References


B 2  1  Abdo, A. A., Ackermann, M., Ajello, M., et al. 2010, Science, 327, 1103
Dp 2  Achterberg, A. 1981, A&A, 97, 259
= 1 2
C(s) d
dx G(x) Antecki, T., Schlickeiser, R., & Zhang, M. 2013, ApJ, 764, 89
p2 4R B0 1 0
 Belcher, J. W., & Davis, L. 1971, J. Geophys. Res., 76, 3534
(R+ + R ) , (B.8a) Bieber, J. W., Matthaeus, W. H., Smith, C. W., et al. 1994, ApJ, 420, 294
Blandford, R. D., & Eichler, D. 1987, Phys. Rep., 154, 1
j=1 Bruno, R., & Carbone, V. 2005, Liv. Rev. Sol. Phys., 2, 1
Bsching, I., Kopp, A., Pohl, M., et al. 2005, ApJ, 619, 314
where Buts, V. A., Manuilenko, O. V., Tolstoluzhsky, A. P., & Turkin, Y. A. 2001, Prob.
  2   Atomic Sci. Tech., 2001, 92
B0 2 B0 1 2 Chandrasekhar, S. 1943, Rev. Mod. Phys., 15, 1
R = x exp x ( j)x R (B.8b)
B B Dosch, A., & Shalchi, A. 2010, Adv. Space Res., 46, 1208

A101, page 9 of 10
A&A 555, A101 (2013)

Drge, W. 2005, Adv. Space Res., 35, 532 Ostrowski, M., & Siemieniec-Oziebo, G. 1997, Astropart. Phys., 6, 271
Drury, L. O. 1983, Rep. Prog. Phys., 46, 973 OSullivan, S., Reville, B., & Taylor, A. M. 2009, MNRAS, 400, 248
Duy, P., & Blundell, K. M. 2005, Plasma Phys. Contr. Fusion, 47, B667 Park, J., & Petrosian, V. 1995, ApJ, 446, 699
Fahr, H., & Fichtner, H. 2011, A&A, 533, A92 Parker, E. N. 1958, ApJ, 128, 664
Fatuzzo, M., & Melia, F. 2011, MNRAS, 410, L23 Parker, E. N. 1965, Planet. Space Sci., 13, 9
Fatuzzo, M., & Melia, F. 2012, ApJ, 750, 21 Parker, E. N., & Tidman, D. A. 1958, Phys. Rev., 111, 1206
Fedorov, Y. I., & Stehlik, M. 2008, in Proc. 21st European Cosmic Ray Pohl, M. 2009, Phys. Rev. D, 79, 041301
Symposium, eds. P. Kirly, K. Kudela, M. Stehlk, & A. W. Wolfendale Press, W. H., Teukolsky, S. A., Vetterling, W. T., & Flannery, B. P. 2007,
(Koice, Slovakia: Inst. Exp. Phys., Slovak Academy of Sciences), 241 Numerical Recipes (Cambridge: University Press)
Fermi, E. 1949, Phys. Rev., 75, 1169 Rausch, M., & Tautz, R. C. 2012, MNRAS, 428, 2333
Fisk, L. A. 1996, J. Geophys. Res., 101, 15547 Schaefer-Rols, U., Lerche, I., & Tautz, R. C. 2009, J. Phys. A: Math. Theor.,
Fisk, L. A., & Gloeckler, G. 2006, ApJ, 640, L79 42, 105501
Fisk, L. A., & Gloeckler, G. 2007a, Space Sci. Rev., 130, 153 Schlickeiser, R. 1985, A&A, 143, 431
Fisk, L. A., & Gloeckler, G. 2007b, Proc. Nat. Acad. Sci., 104, 5749 Schlickeiser, R. 1989a, ApJ, 336, 243
Fisk, L. A., & Gloeckler, G. 2008, ApJ, 686, 1466 Schlickeiser, R. 1989b, ApJ, 336, 246
Fisk, L. A., Gloeckler, G., & Schwadron, N. A. 2010, ApJ, 720, 533 Schlickeiser, R. 1994, ApJS, 20, 929
Giacalone, J., & Jokipii, J. R. 1999, ApJ, 520, 204 Schlickeiser, R. 2002, Cosmic Ray Astrophysics (Berlin: Springer)
Guyer, J., Wheeler, D., & Warren, J. A. 2009, Comp. Sci. Eng., 11, 6 Schlickeiser, R., & Lerche, I. 2007, A&A, 476, 1
Hamilton, R. J., & Petrosian, V. 1992, ApJ, 398, 350 Schlickeiser, R., Artmann, S., & Drge, W. 2009, Open Plasma Phys. J., 2, 1
Hartquist, T. W., & Morfill, G. E. 1983, ApJ, 266, 271 Shalchi, A. 2005, Phys. Plasmas, 12, 052905
Heinbach, U., & Simon, M. 1995, ApJ, 441, 209 Shalchi, A. 2006, MNRAS, 371, 1898
Jokipii, J. R. 1966, ApJ, 146, 480 Shalchi, A., & Schlickeiser, R. 2004, A&A, 420, 799
Jokipii, J. R., & Lee, M. A. 2010, ApJ, 713, 475 Shalchi, A., Bieber, J. W., Matthaeus, W. H., & Qin, G. 2004, ApJ, 616, 617
Klein, M. J. 1955, Phys. Rev. Lett., 97, 1466 Skilling, J. 1975, MNRAS, 172, 557
Kolmogorov, A. N. 1991, Proc. Royal Soc. London, Ser. A: Math. Phys. Sci., Stawarz, ., & Petrosian, V. 2008, ApJ, 681, 1725
434, 9 Tautz, R. C. 2010a, Comput. Phys. Commun., 81, 71
Kuramitsu, Y., Nakanii, N., Kondo, K., et al. 2011, Phys. Plasmas, 18, 010701 Tautz, R. C. 2010b, Plasma Phys. Contr. Fusion, 52, 045016
Lee, M. A., & Vlk, H. J. 1975, ApJ, 198, 485 Tautz, R. C. 2012, in Turbulence: Theory, Types and Simulation, ed. R. J.
Letessier-Selvon, A., & Stanev, T. 2011, Rev. Mod. Phys., 83, 907 Marcuso (New York: Nova Publishers), 365
Litvinenko, Y. E., & Schlickeiser, R. 2011, ApJ, 732, L31 Tautz, R. C., & Lerche, I. 2010, Phys. Lett. A, 374, 4573
Longair, M. S. 2011, High Energy Astrophysics (Cambridge: University Press) Tautz, R. C., & Shalchi, A. 2010, J. Geophys. Res., 115, A03104
Malkov, M. A., Diamond, P., Drury, L. O., & Sagdeev, R. Z. 2010, ApJ, 721, 750 Tautz, R. C., Shalchi, A., & Schlickeiser, R. 2006a, J. Phys. G: Nuclear Part.
Matthaeus, W. H., Goldstein, M. L., & Roberts, D. A. 1990, J. Geophys. Res., Phys., 32, 809
95, 20673 Tautz, R. C., Shalchi, A., & Schlickeiser, R. 2006b, J. Phys. G: Nuclear Part.
Michaek, G., Ostrowski, M., & Schlickeiser, R. 1999, Sol. Phys., 184, Phys., 32, 1045
339 Tautz, R. C., Shalchi, A., & Schlickeiser, R. 2008, ApJ, 685, L165
Moraal, H., Caballero-Lopez, R. A., & Ptuskin, V. S. 2008, in Proc. 3th Tautz, R. C., Dosch, A., & Lerche, I. 2012, A&A, 545, A149
International Cosmic Ray conference, eds. R. Caballero, J. C. DOlivio, Tverskoi, B. A. 1967, J. Exp. Theor. Phys., 25, 317
G. Medina-Tanco, L. Nellen, F. A. Snchez, & J. F. Valds-Galicia (Mexico Vainio, R., & Schlickeiser, R. 1999, A&A, 343, 303
City, Mexico: Universidad Nacional Autnoma de Mxico), 1, 865 Vocks, C., Salem, C., & Lin, R. P. 2005, ApJ, 627, 540
Nagano, M., & Watson, A. A. 2000, Rev. Mod. Phys., 72, 689 Vocks, C., Mann, G., & Rausche, G. 2008, A&A, 480, 527
Ostrowski, M., & Michaek, G. 1993, in 23rd International Cosmic Ray Weinhorst, B., & Shalchi, A. 2010, MNRAS, 403, 287
Conference, held 1930 July, at University of Calgary, Alberta, Canada, Zank, G., Li, G., Florinski, V., et al. 2006, J. Geophys. Res., 111, A06108
eds. D. A. Leahy, R. B. Hickws, & D. Venkatesan (Singapore: World Zhang, M., & Lee, M. A. 2011, Space Sci. Rev.,
Scientific), 2, 309 DOI: 10.1007/s11214-011-9754-3

A101, page 10 of 10

You might also like